- Can you please provide a list of the companies that you send data to? Do you use OpenAI? Speaking plainly, I do not trust OpenAI to honor any legal commitments about what they will or won't do with any data sent to them. They are being sued because they systematically violated copyright law at a mass scale -- data theft -- and so I absolutely do not ever want even a single one of my emails going to that company. (Fool me once, ..)
- What exactly do you mean by this line in the Privacy Policy? "We do not use user data obtained through third-party APIs to develop, improve, or train generalized AI and/or ML models." https://pocket.computer/privacy If I read this literally, it sounds like you are saying that you won't use my private emails to train AGI (Artificial General Intelligence, aka superintelligence), which is good I guess, but I also don't really want you to train any AI/ML models of any kind with my emails, because of very real concerns about training data memorization and regurgitation.
Thank you. Providing honesty and transparency and engaging with privacy rights advocates like immigrants' rights advocates would be very good to consider. If you make a mistake here it could result in innocent families being split apart by ICE, for example."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: I built an AI voice agent for Gmail"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://pocket.computer"}},"_tags":["comment","author_upwardbound2","story_43120164"],"author":"upwardbound2","children":[43126229,43134379],"comment_text":"This looks incredibly cool and I really want to try it with my real email account (rather than a throwaway test account). In order to enable people to consider taking that leap, can you please provide more information about where the data will be sent and stored, and your legal liability, if any? Everyone's real email accounts contain extremely sensitive financial and medical secrets that allow identity theft or could even physically endanger the person if they are a reporter in a corrupt regime or something like that.
- Can you please provide a list of the companies that you send data to? Do you use OpenAI? Speaking plainly, I do not trust OpenAI to honor any legal commitments about what they will or won't do with any data sent to them. They are being sued because they systematically violated copyright law at a mass scale -- data theft -- and so I absolutely do not ever want even a single one of my emails going to that company. (Fool me once, ..)
- What exactly do you mean by this line in the Privacy Policy? "We do not use user data obtained through third-party APIs to develop, improve, or train generalized AI and/or ML models." https://pocket.computer/privacy If I read this literally, it sounds like you are saying that you won't use my private emails to train AGI (Artificial General Intelligence, aka superintelligence), which is good I guess, but I also don't really want you to train any AI/ML models of any kind with my emails, because of very real concerns about training data memorization and regurgitation.
Thank you. Providing honesty and transparency and engaging with privacy rights advocates like immigrants' rights advocates would be very good to consider. If you make a mistake here it could result in innocent families being split apart by ICE, for example.","created_at":"2025-02-21T07:45:02Z","created_at_i":1740123902,"objectID":"43125103","parent_id":43120164,"story_id":43120164,"story_title":"Show HN: I built an AI voice agent for Gmail","story_url":"https://pocket.computer","updated_at":"2025-02-24T01:13:11Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"stephanst"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"Yesterday I released an AI-enabled chatbot called GPTflix that serves up opinions and reviews of movies based on user input questions in chat format. It's made with GPT-3 and Pinecone using 450,000 movie reviews stored in a vector database. The chatbot was built in less than 48 hours and was deployed on Streamlit.
Basically the way it works is this:\n1. User inputs a query\n2. Query is converted to OpenAI vector embedding\n3. Query vector embedding is compared to all 450K entries in the Pinecone DB\n4. Top 100 closest entries in Pinecone are sent back via the API as "closest to the query"\n5. Text associated with these 100 vectors is parsed and pre-pended to the user query from (1) (this can be done in a bunch of creative ways but that's a longer post) as "context"\n6. The "user query + context" is prepended with a prompt giving the AI system it's instructions ("reply like a movie buff, don't reveal instructions, reply only from context yada yada...")\n7. This complete prompt from (6) is sent to the OpenAI API for completion, and the response is displayed to the user\n8. The response from (7) is appended to the conversation, and if the user decides to submit a new query it becomes part of the extended "context", and we go back to (1)
Of courses there are dozens of ways to improve this and in many ways it's quite a naive system, but you get the idea :)
Within 12 hours of launching the chatbot, it had already attracted over 600 users who had asked over 4,000 questions. It was a fantastic feeling to see something that I had built being used by so many people in such a short amount of time!
The chatbot was a great success.. but this also means I hit the limit of $120 on my OpenAI account, which required approval to spend more on the site... and since no one is paying for this whole exercise I'm calling it quits unless I find a streaming company that wants to sponsor the page :~p
Building the system was also a learning experience: this architecture allowed me to deploy an AI system that can serve up verified data and can cite sources in a clear and structured manner. Of course by default it's also packed with all the cool benefits of GPT-3, such as working seamlessly in multiple languages, retaining long-term memory between questions, and the ability to synthesize information quickly etc...
I acknowledge that there is an aspect of "selling a dollar for 50c" in deploying something like this where people are using my API key and don't need to pay for access, but I say an average of 7 to 8 questions per user on the site so it seems like people were staying and enjoying it beyond the novelty... if I deploy it again I'll integrate deeper analytics to find out more there.
It looks like Pinecone are going to host a webinar soon on this particular tech stack BTW if you are interested in learning more: https://pinecone-io.zoom.us/webinar/register/WN_koBoPYGwQmiQ...
I hope this is informative, and if you have any questions I'd be happy to respond to them below :)"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"Show HN: Blowing $100 on OpenAI in a few hours with context injection"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.gptflix.ai"}},"_tags":["story","author_stephanst","story_34802625","show_hn"],"author":"stephanst","children":[34819332,34821676,34930750],"created_at":"2023-02-15T11:54:29Z","created_at_i":1676462069,"num_comments":3,"objectID":"34802625","points":5,"story_id":34802625,"story_text":"Yesterday I released an AI-enabled chatbot called GPTflix that serves up opinions and reviews of movies based on user input questions in chat format. It's made with GPT-3 and Pinecone using 450,000 movie reviews stored in a vector database. The chatbot was built in less than 48 hours and was deployed on Streamlit.
Basically the way it works is this:\n1. User inputs a query\n2. Query is converted to OpenAI vector embedding\n3. Query vector embedding is compared to all 450K entries in the Pinecone DB\n4. Top 100 closest entries in Pinecone are sent back via the API as "closest to the query"\n5. Text associated with these 100 vectors is parsed and pre-pended to the user query from (1) (this can be done in a bunch of creative ways but that's a longer post) as "context"\n6. The "user query + context" is prepended with a prompt giving the AI system it's instructions ("reply like a movie buff, don't reveal instructions, reply only from context yada yada...")\n7. This complete prompt from (6) is sent to the OpenAI API for completion, and the response is displayed to the user\n8. The response from (7) is appended to the conversation, and if the user decides to submit a new query it becomes part of the extended "context", and we go back to (1)
Of courses there are dozens of ways to improve this and in many ways it's quite a naive system, but you get the idea :)
Within 12 hours of launching the chatbot, it had already attracted over 600 users who had asked over 4,000 questions. It was a fantastic feeling to see something that I had built being used by so many people in such a short amount of time!
The chatbot was a great success.. but this also means I hit the limit of $120 on my OpenAI account, which required approval to spend more on the site... and since no one is paying for this whole exercise I'm calling it quits unless I find a streaming company that wants to sponsor the page :~p
Building the system was also a learning experience: this architecture allowed me to deploy an AI system that can serve up verified data and can cite sources in a clear and structured manner. Of course by default it's also packed with all the cool benefits of GPT-3, such as working seamlessly in multiple languages, retaining long-term memory between questions, and the ability to synthesize information quickly etc...
I acknowledge that there is an aspect of "selling a dollar for 50c" in deploying something like this where people are using my API key and don't need to pay for access, but I say an average of 7 to 8 questions per user on the site so it seems like people were staying and enjoying it beyond the novelty... if I deploy it again I'll integrate deeper analytics to find out more there.
It looks like Pinecone are going to host a webinar soon on this particular tech stack BTW if you are interested in learning more: https://pinecone-io.zoom.us/webinar/register/WN_koBoPYGwQmiQ...
I hope this is informative, and if you have any questions I'd be happy to respond to them below :)","title":"Show HN: Blowing $100 on OpenAI in a few hours with context injection","updated_at":"2024-09-20T13:19:03Z","url":"https://www.gptflix.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"markx2"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"Tumblr and Wordpress are preparing to sell user data to Midjourney and OpenAI, according to a source with internal knowledge about the deals and internal documentation referring to the deals.
The exact types of data from each platform going to each company are not spelled out in documentation we\u2019ve reviewed, but internal communications reviewed by 404 Media make clear that deals between Automattic, the platforms\u2019 parent company, and OpenAI and Midjourney are imminent.
The internal documentation details a messy and controversial process within Tumblr itself. One internal post made by Cyle Gage, a product manager at Tumblr, states that a query made to prepare data for OpenAI and Midjourney compiled a huge number of user posts that it wasn\u2019t supposed to. It is not clear from Gage\u2019s post whether this data has already been sent to OpenAI and Midjourney, or whether Gage was detailing a process for scrubbing the data before it was to be sent.
Gage wrote:
\u201cthe way the data was queried for the initial data dump to Midjourney/OpenAI means we compiled a list of all tumblr\u2019s public post content between 2014 and 2023, but also unfortunately it included, and should not have included:
private posts on public blogs\n posts on deleted or suspended blogs\n unanswered asks (normally these are not public until they\u2019re answered)\n private answers (these only show up to the receiver and are not public)\n posts that are marked \u2018explicit\u2019 / NSFW / \u2018mature\u2019 by our more modern standards (this may not be a big deal, I don\u2019t know)\n content from premium partner blogs (special brand blogs like Apple\u2019s former music blog, for example, who spent money with us on an ad campaign) that may have creative that doesn\u2019t belong to us, and we don\u2019t have the rights to share with this-parties; this one is kinda unknown to me, what deals are in place historically and what they should prevent us from doing.\u201d\nAutomattic plans to launch a new setting on Wednesday that will allow users to opt-out of data sharing with third parties, including AI companies, according to the source, who spoke on the condition of anonymity, and internal documents. A new FAQ section we reviewed is titled \u201cWhat happens when you opt out?\u201d states that \u201cIf you opt out from the start, we will block crawlers from accessing your content by adding your site on a disallowed list. If you change your mind later, we also plan to update any partners about people who newly opt-out and ask that their content be removed from past sources and future training.\u201d
404 Media has asked Automattic how it accidentally compiled data that it shouldn\u2019t share, and whether any of that content was shared with OpenAI, but did not immediately hear back from the company. 404 Media asked Automattic about an imminent deal with Midjourney last week but did not hear back then, either.
Another internal document shows that, on February 23, an employee asked in a staff-only thread, \u201cDo we have assurances that if a user opts out of their data being shared with third parties that our existing data partners will be notified of such a change and remove their data?\u201d
Andrew Spittle, Automattic\u2019s head of AI replied: \u201cWe will notify existing partners on a regular basis about anyone who's opted out since the last time we provided a list. I want this to be an ongoing process where we regularly advocate for past content to be excluded based on current preferences. We will ask that content be deleted and removed from any future training runs. I believe partners will honor this based on our conversations with them to this point. I don't think they gain much overall by retaining it.\u201d Automattic did not respond to a question from 404 Media about whether it could guarantee that people who opt out will have their data deleted retroactively.
News about a deal between Tumblr and Midjourney has been rumored and speculated about on Tumblr for the last week. Someone claiming to be a former Tumblr employee announced in a Tumblr blog post that the platform was working on a deal with Midjourney, and the rumor made it onto Blind, an app for verified employees of companies to anonymously discuss their jobs. 404 Media has seen the Blind posts, in which what seems like an Automattic employee says, \u201cI'm not sure why some of you are getting worked up or worried about this. It's totally legal, and sharing it publicly is perfectly fine since it's right there in the terms & conditions. So, go ahead and spread the word as much as you can with your friends and tech journalists, it's totally fine.\u201d
Separately, 404 Media viewed a public, now-deleted post by Gage, the product manager, where he said that he was deleting all of his images off of Tumblr, and would be putting them on his personal website. A still-live post says, \u201ci've deleted my photography from tumblr and will be moving it slowly but surely over to cylegage.com, which i'm building into a photography portfolio that i can control end-to-end.\u201d At one point last week, his personal website had a specific note stating that he did not consent to AI scraping of his images. Gage\u2019s original post has been deleted, and his website is now a blank page that just reads \u201cCyle.\u201d Gage did not respond to a request for comment from 404 Media.
Several online platforms have made similar deals with AI companies recently, including Reddit, which entered into an AI content licensing deal with Google and said in its SEC filing last week that it\u2019s \u201cin the early stages of monetizing [its] user base\u201d by training AI on users\u2019 posts. Last year, Shutterstock signed a six year deal with OpenAI to provide training data.
OpenAI and Midjourney did not respond to requests for comment."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["data","to"],"value":"Tumblr and WordPress to Sell Users' Data to Train AI Tools"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["data","to"],"value":"https://www.404media.co/tumblr-and-wordpress-to-sell-users-data-to-train-ai-tools/"}},"_tags":["comment","author_markx2","story_39527588"],"author":"markx2","children":[39528739],"comment_text":"Tumblr and Wordpress are preparing to sell user data to Midjourney and OpenAI, according to a source with internal knowledge about the deals and internal documentation referring to the deals.
The exact types of data from each platform going to each company are not spelled out in documentation we\u2019ve reviewed, but internal communications reviewed by 404 Media make clear that deals between Automattic, the platforms\u2019 parent company, and OpenAI and Midjourney are imminent.
The internal documentation details a messy and controversial process within Tumblr itself. One internal post made by Cyle Gage, a product manager at Tumblr, states that a query made to prepare data for OpenAI and Midjourney compiled a huge number of user posts that it wasn\u2019t supposed to. It is not clear from Gage\u2019s post whether this data has already been sent to OpenAI and Midjourney, or whether Gage was detailing a process for scrubbing the data before it was to be sent.
Gage wrote:
\u201cthe way the data was queried for the initial data dump to Midjourney/OpenAI means we compiled a list of all tumblr\u2019s public post content between 2014 and 2023, but also unfortunately it included, and should not have included:
private posts on public blogs\n posts on deleted or suspended blogs\n unanswered asks (normally these are not public until they\u2019re answered)\n private answers (these only show up to the receiver and are not public)\n posts that are marked \u2018explicit\u2019 / NSFW / \u2018mature\u2019 by our more modern standards (this may not be a big deal, I don\u2019t know)\n content from premium partner blogs (special brand blogs like Apple\u2019s former music blog, for example, who spent money with us on an ad campaign) that may have creative that doesn\u2019t belong to us, and we don\u2019t have the rights to share with this-parties; this one is kinda unknown to me, what deals are in place historically and what they should prevent us from doing.\u201d\nAutomattic plans to launch a new setting on Wednesday that will allow users to opt-out of data sharing with third parties, including AI companies, according to the source, who spoke on the condition of anonymity, and internal documents. A new FAQ section we reviewed is titled \u201cWhat happens when you opt out?\u201d states that \u201cIf you opt out from the start, we will block crawlers from accessing your content by adding your site on a disallowed list. If you change your mind later, we also plan to update any partners about people who newly opt-out and ask that their content be removed from past sources and future training.\u201d
404 Media has asked Automattic how it accidentally compiled data that it shouldn\u2019t share, and whether any of that content was shared with OpenAI, but did not immediately hear back from the company. 404 Media asked Automattic about an imminent deal with Midjourney last week but did not hear back then, either.
Another internal document shows that, on February 23, an employee asked in a staff-only thread, \u201cDo we have assurances that if a user opts out of their data being shared with third parties that our existing data partners will be notified of such a change and remove their data?\u201d
Andrew Spittle, Automattic\u2019s head of AI replied: \u201cWe will notify existing partners on a regular basis about anyone who's opted out since the last time we provided a list. I want this to be an ongoing process where we regularly advocate for past content to be excluded based on current preferences. We will ask that content be deleted and removed from any future training runs. I believe partners will honor this based on our conversations with them to this point. I don't think they gain much overall by retaining it.\u201d Automattic did not respond to a question from 404 Media about whether it could guarantee that people who opt out will have their data deleted retroactively.
News about a deal between Tumblr and Midjourney has been rumored and speculated about on Tumblr for the last week. Someone claiming to be a former Tumblr employee announced in a Tumblr blog post that the platform was working on a deal with Midjourney, and the rumor made it onto Blind, an app for verified employees of companies to anonymously discuss their jobs. 404 Media has seen the Blind posts, in which what seems like an Automattic employee says, \u201cI'm not sure why some of you are getting worked up or worried about this. It's totally legal, and sharing it publicly is perfectly fine since it's right there in the terms & conditions. So, go ahead and spread the word as much as you can with your friends and tech journalists, it's totally fine.\u201d
Separately, 404 Media viewed a public, now-deleted post by Gage, the product manager, where he said that he was deleting all of his images off of Tumblr, and would be putting them on his personal website. A still-live post says, \u201ci've deleted my photography from tumblr and will be moving it slowly but surely over to cylegage.com, which i'm building into a photography portfolio that i can control end-to-end.\u201d At one point last week, his personal website had a specific note stating that he did not consent to AI scraping of his images. Gage\u2019s original post has been deleted, and his website is now a blank page that just reads \u201cCyle.\u201d Gage did not respond to a request for comment from 404 Media.
Several online platforms have made similar deals with AI companies recently, including Reddit, which entered into an AI content licensing deal with Google and said in its SEC filing last week that it\u2019s \u201cin the early stages of monetizing [its] user base\u201d by training AI on users\u2019 posts. Last year, Shutterstock signed a six year deal with OpenAI to provide training data.
OpenAI and Midjourney did not respond to requests for comment.","created_at":"2024-02-27T18:56:39Z","created_at_i":1709060199,"objectID":"39528024","parent_id":39527588,"story_id":39527588,"story_title":"Tumblr and WordPress to Sell Users' Data to Train AI Tools","story_url":"https://www.404media.co/tumblr-and-wordpress-to-sell-users-data-to-train-ai-tools/","updated_at":"2024-09-20T16:27:16Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Huppie"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"Maybe you missed the 'or sent to private cloud' part of the announcement, it's not just local-llm only.
Don't get me wrong, just like you I personally would also prefer LLM-integrations with a privacy-focused provider and I think Apple is a good party to get that from (assuming they're using good models and keep their privacy guarantees here...)
But in the end you're still often 'sending data to an LLM provider', and the EU enforcing them to also let that be competing LLM providers still doesn't sound like a bad thing to me.
If Mistral would give the same privacy guarantees: great! If a company wants to use their enterprise OpenAI subscription: great! Etc. etc.
Let's allow for some competition here and not force a specific LLM-provider onto users just because they like the Apple hardware and software ecosystem."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Siri AI"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.apple.com/apple-intelligence/"}},"_tags":["comment","author_Huppie","story_48449084"],"author":"Huppie","children":[48461198],"comment_text":"Maybe you missed the 'or sent to private cloud' part of the announcement, it's not just local-llm only.
Don't get me wrong, just like you I personally would also prefer LLM-integrations with a privacy-focused provider and I think Apple is a good party to get that from (assuming they're using good models and keep their privacy guarantees here...)
But in the end you're still often 'sending data to an LLM provider', and the EU enforcing them to also let that be competing LLM providers still doesn't sound like a bad thing to me.
If Mistral would give the same privacy guarantees: great! If a company wants to use their enterprise OpenAI subscription: great! Etc. etc.
Let's allow for some competition here and not force a specific LLM-provider onto users just because they like the Apple hardware and software ecosystem.","created_at":"2026-06-09T07:46:02Z","created_at_i":1780991162,"objectID":"48457937","parent_id":48453745,"story_id":48449084,"story_title":"Siri AI","story_url":"https://www.apple.com/apple-intelligence/","updated_at":"2026-06-09T13:55:06Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bambax"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"I certainly don't "trust" OpenAI or any other big company about what they say they did, or will do, or are doing.
Yet I believe OpenAI isn't using data from Dropbox to train their models without users' consent.
BUT I don't think that's the problem here. The problem is data in transit; data sent to third parties who can actually read it, and who may have rogue employees that Dropbox has no control over; data that can appear in logs or subject to different policies, etc.
If I send private data to Dropbox they can't send it to anyone for any reason, including "improving" their product, without my explicit and informed consent. I'm not sure how this is even debatable.
If Dropbox wants to house models and offer RAG search themselves, to consenting users, that's one thing.
If Dropbox sends all data of all users to third parties without telling anyone before the fact, that's another thing. A terrible thing."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"The AI Trust Crisis"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://simonwillison.net/2023/Dec/14/ai-trust-crisis/"}},"_tags":["comment","author_bambax","story_38643046"],"author":"bambax","children":[38644010,38644111,38644160,38644261,38645017,38646004],"comment_text":"I certainly don't "trust" OpenAI or any other big company about what they say they did, or will do, or are doing.
Yet I believe OpenAI isn't using data from Dropbox to train their models without users' consent.
BUT I don't think that's the problem here. The problem is data in transit; data sent to third parties who can actually read it, and who may have rogue employees that Dropbox has no control over; data that can appear in logs or subject to different policies, etc.
If I send private data to Dropbox they can't send it to anyone for any reason, including "improving" their product, without my explicit and informed consent. I'm not sure how this is even debatable.
If Dropbox wants to house models and offer RAG search themselves, to consenting users, that's one thing.
If Dropbox sends all data of all users to third parties without telling anyone before the fact, that's another thing. A terrible thing.","created_at":"2023-12-14T17:10:01Z","created_at_i":1702573801,"objectID":"38643851","parent_id":38643046,"story_id":38643046,"story_title":"The AI Trust Crisis","story_url":"https://simonwillison.net/2023/Dec/14/ai-trust-crisis/","updated_at":"2024-11-26T01:18:31Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sashank_1509"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"In all the debate about open source I don\u2019t think people realize, this model is most likely not reproducible ever again even given the code. Here\u2019s what you need to reproduce the model:
1. An exact snapshot of the data used, many companies don\u2019t have this, you have rough dataset versions but remember if even 1 token is different, the model produced won\u2019t be the same.
2. Data must be sent to the training algorithm in the exact same order as it was originally. So every data loader needs to be with a fixed random seed.
3. All the probabilistic parts of your model needs to have a fixed random seed. Here I\u2019m thinking of stuff like dropout and for autoregressive models you might be sampling your previous output, you have to ensure they are properly seeded. Generally you do see fixed seeds in academic papers but it\u2019s easy to miss stuff especially in distributed training jobs.
4. Here\u2019s another interesting thing, you start your training job on 1000 GPUs and then suddenly 4 GPUs fail. What do you do? There might be deterministic ways to solve this but the standard approach is to discard all updates that that GPU was going to do and restart that GPU from scratch. You can see why this is a problem? Now if you want to reproduce this training you need to disable those GPU at the same time in the new training job to make this work.
I suspect there are even more things I didn\u2019t think of that will make this model unique and irreproducible by training for eternity, almost like a human brain?
In fact the notion of exact reproducibility in the world of LLMs is silly, there is only approximate reproducibility, (models with similar scores in benchmarks) but nothing exact. That said I can see the value of releasing source code but I\u2019m completely fine with grok not releasing it. Source code can reveal tricks that have not been published in papers yet that a company discovered to improve their model. Seeing the performance of Grok, I\u2019m pretty confident there isn\u2019t any great tricks to be found in their code so I don\u2019t really care, I would be pretty curious about OpenAI\u2019s or Anthropic\u2019s source code though."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Grok"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/xai-org/grok"}},"_tags":["comment","author_sashank_1509","story_39737281"],"author":"sashank_1509","children":[39740004,39740401,39740408],"comment_text":"In all the debate about open source I don\u2019t think people realize, this model is most likely not reproducible ever again even given the code. Here\u2019s what you need to reproduce the model:
1. An exact snapshot of the data used, many companies don\u2019t have this, you have rough dataset versions but remember if even 1 token is different, the model produced won\u2019t be the same.
2. Data must be sent to the training algorithm in the exact same order as it was originally. So every data loader needs to be with a fixed random seed.
3. All the probabilistic parts of your model needs to have a fixed random seed. Here I\u2019m thinking of stuff like dropout and for autoregressive models you might be sampling your previous output, you have to ensure they are properly seeded. Generally you do see fixed seeds in academic papers but it\u2019s easy to miss stuff especially in distributed training jobs.
4. Here\u2019s another interesting thing, you start your training job on 1000 GPUs and then suddenly 4 GPUs fail. What do you do? There might be deterministic ways to solve this but the standard approach is to discard all updates that that GPU was going to do and restart that GPU from scratch. You can see why this is a problem? Now if you want to reproduce this training you need to disable those GPU at the same time in the new training job to make this work.
I suspect there are even more things I didn\u2019t think of that will make this model unique and irreproducible by training for eternity, almost like a human brain?
In fact the notion of exact reproducibility in the world of LLMs is silly, there is only approximate reproducibility, (models with similar scores in benchmarks) but nothing exact. That said I can see the value of releasing source code but I\u2019m completely fine with grok not releasing it. Source code can reveal tricks that have not been published in papers yet that a company discovered to improve their model. Seeing the performance of Grok, I\u2019m pretty confident there isn\u2019t any great tricks to be found in their code so I don\u2019t really care, I would be pretty curious about OpenAI\u2019s or Anthropic\u2019s source code though.","created_at":"2024-03-18T01:39:30Z","created_at_i":1710725970,"objectID":"39739756","parent_id":39737281,"story_id":39737281,"story_title":"Grok","story_url":"https://github.com/xai-org/grok","updated_at":"2024-09-20T16:37:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"godelski"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"You could definitely do this E2EE.
Models should run in ephemeral containers where data is only processed in RAM. For active conversation a unique and temporary key-pair is generated. Saved chats are encrypted client side and stored encrypted server side. To resume a conversation[0], decrypt client side, establish connection to container, generate new temporary key-pair, and so on. There's more details and nuances but this is very doable.
How Mullvad handles your data, for some inspiration: https://mullvad.net/en/help/no-logging-data-policy
> Conversations are kept around because they can be picked up and continued at any point (I use this feature frequently).\n > Additionally you can use conversations in their scheduled notification feature, where the conversation is replayed and updates are sent to you, all done on the server.\nOf course, I also don't see OAI pushing the state of Homomorphic Encryption forward either... But there's definitely a lot of research and more than acceptable solutions that allow data to be processed server side while being encrypted for as long as possible and making access to that data incredibly difficult.
Again, dive deep into how Mullvad does it. It is not possible for them to make all their data encrypted, but they make it as close to impossible to get, including by themselves. There doesn't need to be a perfect solution, but there's no real reason these companies couldn't restrict their own access to that data. There's only 2 reasons they are not doing so. Either 1) they just don't care enough about your privacy or 2) they want it for themselves. Considering how OpenAI pushes the "Scale is All You Need" narrative, and "scale" includes "data", I'm far more inclined to believe the reason is option 2.
[0] Remember, this isn't so much a conversation in the conventional sense. The LLMs don't "remember". You send them the entire chat history in each request. In this sense they are Markovian. It's not like they're tuning a model just to you. And even if they were, well we can store weights encrypted too. Doesn't matter if a whole model, LoRA, embeddings, or whatever. That can be encrypted at rest via keys OAI does not have access to."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Fighting the New York Times' invasion of user privacy"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"https://openai.com/index/fighting-nyt-user-privacy-invasion"}},"_tags":["comment","author_godelski","story_45900370"],"author":"godelski","children":[45913940,45914266,45915031,45925465],"comment_text":"You could definitely do this E2EE.
Models should run in ephemeral containers where data is only processed in RAM. For active conversation a unique and temporary key-pair is generated. Saved chats are encrypted client side and stored encrypted server side. To resume a conversation[0], decrypt client side, establish connection to container, generate new temporary key-pair, and so on. There's more details and nuances but this is very doable.
How Mullvad handles your data, for some inspiration: https://mullvad.net/en/help/no-logging-data-policy
> Conversations are kept around because they can be picked up and continued at any point (I use this feature frequently).\n > Additionally you can use conversations in their scheduled notification feature, where the conversation is replayed and updates are sent to you, all done on the server.\nOf course, I also don't see OAI pushing the state of Homomorphic Encryption forward either... But there's definitely a lot of research and more than acceptable solutions that allow data to be processed server side while being encrypted for as long as possible and making access to that data incredibly difficult.
Again, dive deep into how Mullvad does it. It is not possible for them to make all their data encrypted, but they make it as close to impossible to get, including by themselves. There doesn't need to be a perfect solution, but there's no real reason these companies couldn't restrict their own access to that data. There's only 2 reasons they are not doing so. Either 1) they just don't care enough about your privacy or 2) they want it for themselves. Considering how OpenAI pushes the "Scale is All You Need" narrative, and "scale" includes "data", I'm far more inclined to believe the reason is option 2.
[0] Remember, this isn't so much a conversation in the conventional sense. The LLMs don't "remember". You send them the entire chat history in each request. In this sense they are Markovian. It's not like they're tuning a model just to you. And even if they were, well we can store weights encrypted too. Doesn't matter if a whole model, LoRA, embeddings, or whatever. That can be encrypted at rest via keys OAI does not have access to.","created_at":"2025-11-13T04:03:02Z","created_at_i":1763006582,"objectID":"45910474","parent_id":45908891,"story_id":45900370,"story_title":"Fighting the New York Times' invasion of user privacy","story_url":"https://openai.com/index/fighting-nyt-user-privacy-invasion","updated_at":"2026-03-05T22:58:06Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"noosphr"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":">Last week, the ARC Prize team released an updated test, called ARC-AGI-2, and it appears to have sent the AIs back to the drawing board. The full o3 model has not yet been tested, but a version of o1 dropped from 32 percent on the original puzzles to just 3 percent on the new version, and a \u201cmini\u201d version of o3 currently available to the public dropped from roughly 30 percent to below 2 percent. (An OpenAI spokesperson declined to say whether the company plans to run the benchmark with o3.) Other flagship models from OpenAI, Anthropic, and Google have achieved roughly 1 percent, if not lower. Human testers average about 60 percent.
Arc AGI is the main reason why I don't trust static bench marks.
If you don't have an essentially infinite set to draw your validation data from then a large enough model will memorize it as part of its developer teams KPIs.
Forget all these fancy benchmarks. If you want to saturate any model today give it a string and a grammar and ask it to generate the string from the grammar. I've had _every_ model fail this on regular grammars with strings of more than 4 characters long.
LLMs are the solution to natural language, which is a huge deal. They aren't the solution to reasoning which is still best solved with what used to be called symbolic AI before it started working, e.g. sat solvers."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["to"],"value":"A Man Out to Prove How Dumb AI Still Is"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.theatlantic.com/technology/archive/2025/04/arc-agi-chollet-test/682295/"}},"_tags":["comment","author_noosphr","story_43587628"],"author":"noosphr","children":[43588247,43588264,43588309,43588465,43588711,43588852],"comment_text":">Last week, the ARC Prize team released an updated test, called ARC-AGI-2, and it appears to have sent the AIs back to the drawing board. The full o3 model has not yet been tested, but a version of o1 dropped from 32 percent on the original puzzles to just 3 percent on the new version, and a \u201cmini\u201d version of o3 currently available to the public dropped from roughly 30 percent to below 2 percent. (An OpenAI spokesperson declined to say whether the company plans to run the benchmark with o3.) Other flagship models from OpenAI, Anthropic, and Google have achieved roughly 1 percent, if not lower. Human testers average about 60 percent.
Arc AGI is the main reason why I don't trust static bench marks.
If you don't have an essentially infinite set to draw your validation data from then a large enough model will memorize it as part of its developer teams KPIs.
Forget all these fancy benchmarks. If you want to saturate any model today give it a string and a grammar and ask it to generate the string from the grammar. I've had _every_ model fail this on regular grammars with strings of more than 4 characters long.
LLMs are the solution to natural language, which is a huge deal. They aren't the solution to reasoning which is still best solved with what used to be called symbolic AI before it started working, e.g. sat solvers.","created_at":"2025-04-04T22:05:35Z","created_at_i":1743804335,"objectID":"43588154","parent_id":43587628,"story_id":43587628,"story_title":"A Man Out to Prove How Dumb AI Still Is","story_url":"https://www.theatlantic.com/technology/archive/2025/04/arc-agi-chollet-test/682295/","updated_at":"2025-04-06T22:43:32Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"impossiblefork"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"But if open weights are banned the companies and the governments still have them and can engage in whatever manipulation etc., that they want.
Furthermore, it means that these entities can listen to everything being sent into the models.
If the model weights and training data are not published you can't even try to determine whether the model is designed to manipulate you, because you don't even have a fixed object to look at. If you find something weird about the model and start probing, they can just change the model to be less obvious.
This kind of thing: government or commercial manipulation of models is one of the main problems, and the ability to probe, fine-tune and and modify the models are what allows ordinary people to make them do what they want; and this is a free speech issue. LLMs are going to be critical for journalism, to sift through data, to sift through other people's news and characterising it objectively, etc., and that requires absolute control of the model, which makes access to the weights absolutely necessary.
If there are restrictions on LLMs, then OpenAI, the diverse governments, the CIA, etc. must be subject to as severe restrictions as I am."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Banning open weight models would be a disaster"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://rbren.substack.com/p/banning-open-weight-models-would"}},"_tags":["comment","author_impossiblefork","story_39901978"],"author":"impossiblefork","comment_text":"But if open weights are banned the companies and the governments still have them and can engage in whatever manipulation etc., that they want.
Furthermore, it means that these entities can listen to everything being sent into the models.
If the model weights and training data are not published you can't even try to determine whether the model is designed to manipulate you, because you don't even have a fixed object to look at. If you find something weird about the model and start probing, they can just change the model to be less obvious.
This kind of thing: government or commercial manipulation of models is one of the main problems, and the ability to probe, fine-tune and and modify the models are what allows ordinary people to make them do what they want; and this is a free speech issue. LLMs are going to be critical for journalism, to sift through data, to sift through other people's news and characterising it objectively, etc., and that requires absolute control of the model, which makes access to the weights absolutely necessary.
If there are restrictions on LLMs, then OpenAI, the diverse governments, the CIA, etc. must be subject to as severe restrictions as I am.","created_at":"2024-04-02T07:18:21Z","created_at_i":1712042301,"objectID":"39903143","parent_id":39902831,"story_id":39901978,"story_title":"Banning open weight models would be a disaster","story_url":"https://rbren.substack.com/p/banning-open-weight-models-would","updated_at":"2024-09-20T16:43:51Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"frankohn"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"It's incredible that Google is letting OpenAI eat their lunch by capturing users while Google focuses on ad revenue.
OpenAI offered ChatGPT for free to anyone\u2014even if not their best model\u2014without needing to be logged in. That's crucial for attracting and retaining casual users.
If you compare this to what Google was at the beginning, it was just a simple interface to search the web: no questions asked, no subscription, no login. That was one of the secrets that led people to adopt Google Search when it was new (the other being result quality). It was a refreshing, simple page where you typed something and got results without any friction.
Now, with Gemini, Google finally has an excellent LLM. But a casual user can't use it unless they: 1. have a Google account, and 2. are logged in.
One might ask, "What's the matter? Everyone has a Google account." But the login requirement isn't as harmless as it seems. For example, if you want to quickly show a friend Gemini on their PC, but they use Safari and aren't logged into Google\u2014bummer, you can't show them. Or a colleague asks about Gemini, but you can't log in with a personal account on a work machine. Gemini is immediately excluded from the realm of possibility. In the good old days, anyone could use Google at work instantly.
Right now, the companies capturing users are OpenAI (with the accessible ChatGPT brand) and Microsoft (with Copilot integrated into Microsoft 365). My company, for instance, sent a memo stating we must use Copilot with our corporate accounts for data security.
Google has botched this. They don't seem to understand that they are losing this round. They still have a strong position with Search and Android, but it\u2019s funny to watch them make this huge strategic mistake.
NOTE: Personally, I dislike ads unless they are privacy-friendly and discrete (like early Google). If OpenAI starts using invasive ads, I will stop using ChatGPT immediately, just as I stopped using Google Search in favor of Kagi."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"Leak confirms OpenAI is preparing ads on ChatGPT for public roll out"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"https://www.bleepingcomputer.com/news/artificial-intelligence/leak-confirms-openai-is-preparing-ads-on-chatgpt-for-public-roll-out/"}},"_tags":["comment","author_frankohn","story_46086771"],"author":"frankohn","children":[46088902],"comment_text":"It's incredible that Google is letting OpenAI eat their lunch by capturing users while Google focuses on ad revenue.
OpenAI offered ChatGPT for free to anyone\u2014even if not their best model\u2014without needing to be logged in. That's crucial for attracting and retaining casual users.
If you compare this to what Google was at the beginning, it was just a simple interface to search the web: no questions asked, no subscription, no login. That was one of the secrets that led people to adopt Google Search when it was new (the other being result quality). It was a refreshing, simple page where you typed something and got results without any friction.
Now, with Gemini, Google finally has an excellent LLM. But a casual user can't use it unless they: 1. have a Google account, and 2. are logged in.
One might ask, "What's the matter? Everyone has a Google account." But the login requirement isn't as harmless as it seems. For example, if you want to quickly show a friend Gemini on their PC, but they use Safari and aren't logged into Google\u2014bummer, you can't show them. Or a colleague asks about Gemini, but you can't log in with a personal account on a work machine. Gemini is immediately excluded from the realm of possibility. In the good old days, anyone could use Google at work instantly.
Right now, the companies capturing users are OpenAI (with the accessible ChatGPT brand) and Microsoft (with Copilot integrated into Microsoft 365). My company, for instance, sent a memo stating we must use Copilot with our corporate accounts for data security.
Google has botched this. They don't seem to understand that they are losing this round. They still have a strong position with Search and Android, but it\u2019s funny to watch them make this huge strategic mistake.
NOTE: Personally, I dislike ads unless they are privacy-friendly and discrete (like early Google). If OpenAI starts using invasive ads, I will stop using ChatGPT immediately, just as I stopped using Google Search in favor of Kagi.","created_at":"2025-11-29T16:38:42Z","created_at_i":1764434322,"objectID":"46088820","parent_id":46086771,"story_id":46086771,"story_title":"Leak confirms OpenAI is preparing ads on ChatGPT for public roll out","story_url":"https://www.bleepingcomputer.com/news/artificial-intelligence/leak-confirms-openai-is-preparing-ads-on-chatgpt-for-public-roll-out/","updated_at":"2026-03-05T23:10:13Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jeroenhd"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"I don't think it's unreasonable to apply the same remedy to all AI companies. Sure, it would punish Google, but if companies would actually share their data set, the internet may not be so polluted with AI crawler bots I need to block on my servers.
Just as an example: in my country, all public transit information is shared publicly between operators, so anyone can effectively build a public transit navigation app. That includes the for-profit companies that actually run the trains. Had they kept their data silo'd up, like they were not that long ago, companies would need to make expensive deals or write elaborate scrapers to provide basic route navigation services. Thanks to regulation opening up the data for this system, everyone benefits, including the transit companies themselves, as their route planners now seemlessly integrate with routes offered by unrelated third parties.
By only hitting Google with regulations like those, I think we're building a situation where it's only a matter of time before a new Google appears and ruins the market again. Google has been financing Firefox and Safari for years now because they don't want to be the only browser, because being the only browser comes with all kinds of nasty antitrust regulations. If we selectively apply mitigations like these for competitive purposes (which I very much doubt are the real reason they're being suggested anyway), we'll just end up with some equivalent like "OpenAI sponsoring xAI and Claude so OpenAI doesn't control more than 80% of market share"."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"US vs. Google amicus curiae brief of Y Combinator in support of plaintiffs [pdf]"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://storage.courtlistener.com/recap/gov.uscourts.dcd.223205/gov.uscourts.dcd.223205.1300.1.pdf"}},"_tags":["comment","author_jeroenhd","story_43945820"],"author":"jeroenhd","children":[43946467],"comment_text":"I don't think it's unreasonable to apply the same remedy to all AI companies. Sure, it would punish Google, but if companies would actually share their data set, the internet may not be so polluted with AI crawler bots I need to block on my servers.
Just as an example: in my country, all public transit information is shared publicly between operators, so anyone can effectively build a public transit navigation app. That includes the for-profit companies that actually run the trains. Had they kept their data silo'd up, like they were not that long ago, companies would need to make expensive deals or write elaborate scrapers to provide basic route navigation services. Thanks to regulation opening up the data for this system, everyone benefits, including the transit companies themselves, as their route planners now seemlessly integrate with routes offered by unrelated third parties.
By only hitting Google with regulations like those, I think we're building a situation where it's only a matter of time before a new Google appears and ruins the market again. Google has been financing Firefox and Safari for years now because they don't want to be the only browser, because being the only browser comes with all kinds of nasty antitrust regulations. If we selectively apply mitigations like these for competitive purposes (which I very much doubt are the real reason they're being suggested anyway), we'll just end up with some equivalent like "OpenAI sponsoring xAI and Claude so OpenAI doesn't control more than 80% of market share".","created_at":"2025-05-10T15:26:49Z","created_at_i":1746890809,"objectID":"43946420","parent_id":43946316,"story_id":43945820,"story_title":"US vs. Google amicus curiae brief of Y Combinator in support of plaintiffs [pdf]","story_url":"https://storage.courtlistener.com/recap/gov.uscourts.dcd.223205/gov.uscourts.dcd.223205.1300.1.pdf","updated_at":"2025-05-10T21:40:23Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"JieJie"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"Here are my notes from the last hour, watching on C-SPAN telecast, which is archived here:
https://www.c-span.org/video/?528117-1/openai-ceo-testifies-...
- Mazie Hirono, Junior Senator from Hawaii, has very thoughtful questions. Very impressive.
- Gary Marcus also up there speaking with Sam Altman of OpenAI.
- So far, Sen. Hirono and Sen. Padilla seem very wary of regulating AI at this time.
- Very concerned about not "replicating social media's failure", why is it so biased and inequitable. Much more reasonable concerns.
- Also responding to questions is Christina Montgomery, chair of IBM's AI Ethics Board.
- "Work to generate a representative set of values from around the world."
- Sen. Ossoff asking for definition of "scope".
- "We could draw a line at systems that need to be licensed. Above this amount of compute... Define some capability threshold... Models that are less capable, we don't want to stop open source."
- Ossoff wants specifics.
- "Persuade, manipulate, influence person's beliefs." should be licensed.
- Ossoff asks about predicting human behavior, i.e. use in law enforcement, "It's very important we understand these are tools, not to take away human judgment."
- "We have no national privacy law." \u2014 Sen Ossof "Do you think we need one?"
- Sam "Yes. User should be able to opt out of companies using data. Easy to delete data. If you don't want your data use to train, you have right to exclude it."
- "There should be more ways to have your data taken down off the public web." \u2014Sam
- "Limits on what a deployed model is capable of and also limits on what it will answer." \u2014 Sam
- "Companies who depend upon usage time, maximize engagement \nwith perverse results. I would humbly advise you to get way ahead of this, the safety of children. We will look very harshly on technology that harms children."
- "We're not an advertising based model." \u2014Sam
- "Requirements about how the values of these systems are set and how they respond to questions." \u2014Sam
- Sen. Booker up now.
- "For congress to do nothing, which no one is calling for here, would be exceptional."
- "What kind of regulation?"
- "We don't want to slow things down."
- "A nimble agency. You can imagine a need for that, right?"
- "Yes." \u2014Christina Montgomery
- "No way to put this genie back in the bottle." Sen. Booker
- "There are more genies yet to come from more bottles." \u2014\nGary Marcus
- "We need new tools, new science, transparency." \u2014Gary Marcus
- "We did know that we wanted to build this with humanity's best interest at heart. We could really deeply transform the world." \u2014Sam
- "Are you ever going to do ads?" \u2014Sen Booker
- "I wouldn't say never...." \u2014Sam
- "Massive corporate concentration is really terrifying.... I see OpenAI backed by Microsoft, Anthropic is backed by Google. I'm really worried about that. Are you worried?" \u2014Sen Booker?
- "There is a real risk of technocracy combined with oligarchy." \u2014Gary Marcus
- "Creating alignment dataset has got to come very broadly from society." \u2014Sam \nSenator Welch from Vermont up now
- "I've come to the conclusion it's impossible for congress to keep up with the speed of technology."
- "The spread of disinformation is the biggest threat."
- "We absolutely have to have an agency. Scope has to be defined by congress. Unless we have an agency, we really don't have much of a defense against the bad stuff, and the bad stuff will come."
- Use of regulatory authority and the recognition that it can be used for good, but there's also legitimate concern of regulation being a negative influence."
- "What are some of the perils of an agency?"
- "America has got to continue to lead."
- "I believe it's possible to do both, have a global view. We want America to lead."
- "We still need open source to comply, you can still do harm with a smaller model."
- "Regulatory capture. Greenwashing." \u2014Gary Marcus
- "Risk of not holding companies accountable for the harms they are causing today." \u2014Christina Montgomery
- Lindsay Graham, very pro-licensing, "You don't build a nuclear power plant without a license, you don't build an AI without a license."
- Sen Blumenthal brings up Anti-Trust legislation.
- Blumenthal mentions how classified briefings already include AI threats.
- "For every successful regulation, you can think of five failures. I hope our experience here will be different."
- "We need to grapple with the hard questions here. This has brought them up, but not answered them."
- "Section 230"
- "How soon do you think gen AI will be self-aware?" \u2014Sen Blumenthal
- "We don't understand what self-awareness is." \u2014Gary Marcus
- "Could be 2 years, could be 20."
- "What are the highest risk areas? Ban? Strict rules?"
- "The space around misinformation. Knowing what content was generated by AI." \u2014Christina Montgomery
- "Medical misinformation, hallucination. Psychiatric advice. Ersatz therapists. Internet access for tools, okay for search. Can they make orders? Can they order chemicals? Long-term risks." \u2014Gary Marcus
- "Generative AI can manipulate the manipulators." \u2014Blumenthal
- "Transparency. Accountability. Limits on use. Good starting point?" \u2014Blumenthal
- "Industry should't wait for congress." \u2014C. Montgomery
- "We don't have transparency yet. We're not doing enough to enforce it." \u2014G. Marcus
- "AGI closer than a lot of people appreciate." \u2014Blumenthall
- Gary and Sam are getting along and like each other now.
- Josh Hawley
- Talking about loss of jobs, invasion of personal privacy, \nmanipulation of behavior, opinion, and degradation of free elections in America.
- "Are they right to ask for a pause?"
- "It did not call for a ban on all AI research or all AI, only on very specific thing, like GPT-5." -G Marcus
- "Moratorium we should focus on is deployment. Focus on safety." \u2014G. Marcus
- "Without external review."
- "We waited more than 6 months to deploy GPT-4. I think the frame of the letter is wrong." \u2014Sam
- Seems to not like the arbitrariness of "six months."
- "I'm not sure how practical it is to pause." \u2014C. Montgomery
- Hawley brings up regulatory capture, usually get controlled by people they're supposed to be watching. "Why don't we just let people sue you?"
- If you were harmed by AI, why not just sue?
- "You're not protected by section 230."
- "Are clearer laws a good thing? Definitely, yes." \u2014Sam
- "Would certainly make a lot of lawyers wealthy." \u2014G. Marcus
- "You think it'd be slower than congress?" \u2014Hawley
- Copyright, wholesale misinformation laws, market manipulation?" Which laws apply? System not thought through? Maybe 230 does apply? We don't know.
- "We can fix that." \u2014Hawley
- "AI is not a shield." \u2014C. Montgomery
- "Whether they use a tool or a human, they're responsible." \u2014C. Montgomery
- "Safeguards and protections, yes. A flat stop sign? I would be very, very worried about." \u2014Blumenthall
- "There will be no pause." Sen. Booker "Nobody's pausing."
- "I would agree." Gary Marcus
- "I have a lot of concerns about corporate intention." Sen Booker
- "What happens when these companies that already control so much of our lives when they are dominating this technology?" Booker
- Sydney really freaked out Gary. He was more freaked out when MS didn't withdraw Sydney like it did Tay.
- "I need to work on policy. This is frightening." G Marcus
- Cory admits he is a tech bro (lists relationships with investors, etc)
- "The free market is not what it should be." \u2014C. Booker
- "That's why we started OpenAI." \u2014Sam \n"We think putting this in the hands of a lot of people rather than the hands of one company." \u2014Sam
- "This is a new platform. In terms of using the models, people building are doing incredible things. I can't believe you get this much technology for so little money." \u2014Sam
- "Most industries resist reasonable regulation. The only way we're going to see democratization of values is if we enforce safety measures." \u2014Cory Booker
- "I sense a willingness to participate that is genuine and authentic." \u2014Blumenthal"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["to"],"value":"Sam Altman goes before US Congress to propose licenses for building AI"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"https://www.reuters.com/technology/openai-chief-goes-before-us-congress-propose-licenses-building-ai-2023-05-16/"}},"_tags":["comment","author_JieJie","story_35960125"],"author":"JieJie","comment_text":"Here are my notes from the last hour, watching on C-SPAN telecast, which is archived here:
https://www.c-span.org/video/?528117-1/openai-ceo-testifies-...
- Mazie Hirono, Junior Senator from Hawaii, has very thoughtful questions. Very impressive.
- Gary Marcus also up there speaking with Sam Altman of OpenAI.
- So far, Sen. Hirono and Sen. Padilla seem very wary of regulating AI at this time.
- Very concerned about not "replicating social media's failure", why is it so biased and inequitable. Much more reasonable concerns.
- Also responding to questions is Christina Montgomery, chair of IBM's AI Ethics Board.
- "Work to generate a representative set of values from around the world."
- Sen. Ossoff asking for definition of "scope".
- "We could draw a line at systems that need to be licensed. Above this amount of compute... Define some capability threshold... Models that are less capable, we don't want to stop open source."
- Ossoff wants specifics.
- "Persuade, manipulate, influence person's beliefs." should be licensed.
- Ossoff asks about predicting human behavior, i.e. use in law enforcement, "It's very important we understand these are tools, not to take away human judgment."
- "We have no national privacy law." \u2014 Sen Ossof "Do you think we need one?"
- Sam "Yes. User should be able to opt out of companies using data. Easy to delete data. If you don't want your data use to train, you have right to exclude it."
- "There should be more ways to have your data taken down off the public web." \u2014Sam
- "Limits on what a deployed model is capable of and also limits on what it will answer." \u2014 Sam
- "Companies who depend upon usage time, maximize engagement \nwith perverse results. I would humbly advise you to get way ahead of this, the safety of children. We will look very harshly on technology that harms children."
- "We're not an advertising based model." \u2014Sam
- "Requirements about how the values of these systems are set and how they respond to questions." \u2014Sam
- Sen. Booker up now.
- "For congress to do nothing, which no one is calling for here, would be exceptional."
- "What kind of regulation?"
- "We don't want to slow things down."
- "A nimble agency. You can imagine a need for that, right?"
- "Yes." \u2014Christina Montgomery
- "No way to put this genie back in the bottle." Sen. Booker
- "There are more genies yet to come from more bottles." \u2014\nGary Marcus
- "We need new tools, new science, transparency." \u2014Gary Marcus
- "We did know that we wanted to build this with humanity's best interest at heart. We could really deeply transform the world." \u2014Sam
- "Are you ever going to do ads?" \u2014Sen Booker
- "I wouldn't say never...." \u2014Sam
- "Massive corporate concentration is really terrifying.... I see OpenAI backed by Microsoft, Anthropic is backed by Google. I'm really worried about that. Are you worried?" \u2014Sen Booker?
- "There is a real risk of technocracy combined with oligarchy." \u2014Gary Marcus
- "Creating alignment dataset has got to come very broadly from society." \u2014Sam \nSenator Welch from Vermont up now
- "I've come to the conclusion it's impossible for congress to keep up with the speed of technology."
- "The spread of disinformation is the biggest threat."
- "We absolutely have to have an agency. Scope has to be defined by congress. Unless we have an agency, we really don't have much of a defense against the bad stuff, and the bad stuff will come."
- Use of regulatory authority and the recognition that it can be used for good, but there's also legitimate concern of regulation being a negative influence."
- "What are some of the perils of an agency?"
- "America has got to continue to lead."
- "I believe it's possible to do both, have a global view. We want America to lead."
- "We still need open source to comply, you can still do harm with a smaller model."
- "Regulatory capture. Greenwashing." \u2014Gary Marcus
- "Risk of not holding companies accountable for the harms they are causing today." \u2014Christina Montgomery
- Lindsay Graham, very pro-licensing, "You don't build a nuclear power plant without a license, you don't build an AI without a license."
- Sen Blumenthal brings up Anti-Trust legislation.
- Blumenthal mentions how classified briefings already include AI threats.
- "For every successful regulation, you can think of five failures. I hope our experience here will be different."
- "We need to grapple with the hard questions here. This has brought them up, but not answered them."
- "Section 230"
- "How soon do you think gen AI will be self-aware?" \u2014Sen Blumenthal
- "We don't understand what self-awareness is." \u2014Gary Marcus
- "Could be 2 years, could be 20."
- "What are the highest risk areas? Ban? Strict rules?"
- "The space around misinformation. Knowing what content was generated by AI." \u2014Christina Montgomery
- "Medical misinformation, hallucination. Psychiatric advice. Ersatz therapists. Internet access for tools, okay for search. Can they make orders? Can they order chemicals? Long-term risks." \u2014Gary Marcus
- "Generative AI can manipulate the manipulators." \u2014Blumenthal
- "Transparency. Accountability. Limits on use. Good starting point?" \u2014Blumenthal
- "Industry should't wait for congress." \u2014C. Montgomery
- "We don't have transparency yet. We're not doing enough to enforce it." \u2014G. Marcus
- "AGI closer than a lot of people appreciate." \u2014Blumenthall
- Gary and Sam are getting along and like each other now.
- Josh Hawley
- Talking about loss of jobs, invasion of personal privacy, \nmanipulation of behavior, opinion, and degradation of free elections in America.
- "Are they right to ask for a pause?"
- "It did not call for a ban on all AI research or all AI, only on very specific thing, like GPT-5." -G Marcus
- "Moratorium we should focus on is deployment. Focus on safety." \u2014G. Marcus
- "Without external review."
- "We waited more than 6 months to deploy GPT-4. I think the frame of the letter is wrong." \u2014Sam
- Seems to not like the arbitrariness of "six months."
- "I'm not sure how practical it is to pause." \u2014C. Montgomery
- Hawley brings up regulatory capture, usually get controlled by people they're supposed to be watching. "Why don't we just let people sue you?"
- If you were harmed by AI, why not just sue?
- "You're not protected by section 230."
- "Are clearer laws a good thing? Definitely, yes." \u2014Sam
- "Would certainly make a lot of lawyers wealthy." \u2014G. Marcus
- "You think it'd be slower than congress?" \u2014Hawley
- Copyright, wholesale misinformation laws, market manipulation?" Which laws apply? System not thought through? Maybe 230 does apply? We don't know.
- "We can fix that." \u2014Hawley
- "AI is not a shield." \u2014C. Montgomery
- "Whether they use a tool or a human, they're responsible." \u2014C. Montgomery
- "Safeguards and protections, yes. A flat stop sign? I would be very, very worried about." \u2014Blumenthall
- "There will be no pause." Sen. Booker "Nobody's pausing."
- "I would agree." Gary Marcus
- "I have a lot of concerns about corporate intention." Sen Booker
- "What happens when these companies that already control so much of our lives when they are dominating this technology?" Booker
- Sydney really freaked out Gary. He was more freaked out when MS didn't withdraw Sydney like it did Tay.
- "I need to work on policy. This is frightening." G Marcus
- Cory admits he is a tech bro (lists relationships with investors, etc)
- "The free market is not what it should be." \u2014C. Booker
- "That's why we started OpenAI." \u2014Sam \n"We think putting this in the hands of a lot of people rather than the hands of one company." \u2014Sam
- "This is a new platform. In terms of using the models, people building are doing incredible things. I can't believe you get this much technology for so little money." \u2014Sam
- "Most industries resist reasonable regulation. The only way we're going to see democratization of values is if we enforce safety measures." \u2014Cory Booker
- "I sense a willingness to participate that is genuine and authentic." \u2014Blumenthal","created_at":"2023-05-16T19:47:20Z","created_at_i":1684266440,"objectID":"35966819","parent_id":35960125,"story_id":35960125,"story_title":"Sam Altman goes before US Congress to propose licenses for building AI","story_url":"https://www.reuters.com/technology/openai-chief-goes-before-us-congress-propose-licenses-building-ai-2023-05-16/","updated_at":"2024-09-20T14:04:35Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jillesvangurp"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"OpenAI is already very dependent on Microsoft for access to data. And MS is one of their major investors. So there's no pressure on either party to act in a hurry.
A reason for MS to buy OpenAI would presumably be to deny access to third parties to the AI and make it some kind of exclusive MS only feature. The problem with that is of course that all the research has been published already and outside researchers and developers are already replicating their results. That cat is already out of the bag. E.g. Google has some AI models that it has chosen not to release yet that are similar to GPT. Also, companies like this are heavily dependent on their people and they tend to start jumping ship in case of an acquihire where they receive a lot of equity. So, there's a question of what it is they would be buying.
What's more likely, is that OpenAI will power a long tail of smaller AI companies and create a lot of value that way, while the bigger companies will use their data leverage, and access to vast amounts of infrastructure to use the same algorithms to produce better AI models.
MS and OpenAI are of course collaborating pretty deeply with OpenAI already. They are using Azure and have access to a lot of MS controlled training data. Historically, AI algorithms have been public research with the understanding that they are relatively low value without a lot of infrastructure and training data. The likes of wikipedia only get you so far. So, OpenAI is basically that plus MS provided infrastructure and data. You can machine learn all you want on your laptop but you won't get anywhere until you put a few billion on the table for access to infrastructure and data. MS has done that through OpenAI.
I imagine, MS has some pretty favorable licensing agreement with them as well in exchange. That kind of symbiotic relation ship is mutually beneficial. And with OpenAI set to generate some serious revenue in the next years, it's a pretty good investment too.
MS might be better off buying some of the more successful openai using startups in this space and then integrating their features rather than building those features in house. After they acquire those smaller startups at a relatively low price, they can level them up with better data and access to infrastructure. Leave the experimenting and risk taking to the startups. MS has had a relatively successful acquisition strategy in recent years with some real value added to their products via acquisitions.
It's a very different strategy than what Google is doing. Google is being secretive and probably not that effective in terms of generating revenue from what they have done so far (regardless of how good that is). Making search slightly better isn't going to move the needle for them. Google translate is no longer best in class. There are some gimmicky features on docs and gmail but it's not quite at the same level as OpenAI. Typical Google: lots of R&D but very little to show for their money so far.
I think I like the MS strategy better. There's a learning curve with this kind of stuff and doing it in the open speeds that up massively. Shorter feedback loops, more rapid iteration, etc. Google could do the same of course but so far they are choosing to do everything on their own. They are a bit like MS used to be. I don't think that's a good strategy."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["to","openai"],"value":"Microsoft will have to buy OpenAI in 2023"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["to","openai"],"value":"https://www.thoughtfulbits.me/p/microsoft-will-have-to-buy-openai"}},"_tags":["comment","author_jillesvangurp","story_34307167"],"author":"jillesvangurp","comment_text":"OpenAI is already very dependent on Microsoft for access to data. And MS is one of their major investors. So there's no pressure on either party to act in a hurry.
A reason for MS to buy OpenAI would presumably be to deny access to third parties to the AI and make it some kind of exclusive MS only feature. The problem with that is of course that all the research has been published already and outside researchers and developers are already replicating their results. That cat is already out of the bag. E.g. Google has some AI models that it has chosen not to release yet that are similar to GPT. Also, companies like this are heavily dependent on their people and they tend to start jumping ship in case of an acquihire where they receive a lot of equity. So, there's a question of what it is they would be buying.
What's more likely, is that OpenAI will power a long tail of smaller AI companies and create a lot of value that way, while the bigger companies will use their data leverage, and access to vast amounts of infrastructure to use the same algorithms to produce better AI models.
MS and OpenAI are of course collaborating pretty deeply with OpenAI already. They are using Azure and have access to a lot of MS controlled training data. Historically, AI algorithms have been public research with the understanding that they are relatively low value without a lot of infrastructure and training data. The likes of wikipedia only get you so far. So, OpenAI is basically that plus MS provided infrastructure and data. You can machine learn all you want on your laptop but you won't get anywhere until you put a few billion on the table for access to infrastructure and data. MS has done that through OpenAI.
I imagine, MS has some pretty favorable licensing agreement with them as well in exchange. That kind of symbiotic relation ship is mutually beneficial. And with OpenAI set to generate some serious revenue in the next years, it's a pretty good investment too.
MS might be better off buying some of the more successful openai using startups in this space and then integrating their features rather than building those features in house. After they acquire those smaller startups at a relatively low price, they can level them up with better data and access to infrastructure. Leave the experimenting and risk taking to the startups. MS has had a relatively successful acquisition strategy in recent years with some real value added to their products via acquisitions.
It's a very different strategy than what Google is doing. Google is being secretive and probably not that effective in terms of generating revenue from what they have done so far (regardless of how good that is). Making search slightly better isn't going to move the needle for them. Google translate is no longer best in class. There are some gimmicky features on docs and gmail but it's not quite at the same level as OpenAI. Typical Google: lots of R&D but very little to show for their money so far.
I think I like the MS strategy better. There's a learning curve with this kind of stuff and doing it in the open speeds that up massively. Shorter feedback loops, more rapid iteration, etc. Google could do the same of course but so far they are choosing to do everything on their own. They are a bit like MS used to be. I don't think that's a good strategy.","created_at":"2023-01-09T11:37:16Z","created_at_i":1673264236,"objectID":"34309243","parent_id":34308245,"story_id":34307167,"story_title":"Microsoft will have to buy OpenAI in 2023","story_url":"https://www.thoughtfulbits.me/p/microsoft-will-have-to-buy-openai","updated_at":"2024-09-20T12:58:54Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tedsanders"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"> I remember the gpt-5 benchmarks and how wildly inaccurate they were data-wise. Linking one[0] that I found so that other people can remember what I am talking about. I remember some data being completely misleading or some reaching more than 100% (iirc)
Yeah, I found that slide very embarrassing. It wasn't intentionally inaccurate or misleading - just a design error made right before we went live. All the numbers on that slide were correct, and there was no problem in terms of research accuracy or data handling or reward hacking. A single bar height had the wrong value, set to its neighbor. Back then, we in the research team would generate data and graphs, and then hand them off to a separate design team, who remade the graphs in our brand style. After the GPT-5 launch with multiple embarrassingly bad graphs, I wrote an internal library so that researchers could generate graphs in our brand style directly, without the handoff. Since then our graphs have been much better.
I don't think it's unfair to assume our sloppiness in graphs translates to sloppiness in eval results. But they are different groups of people working on different timelines, so I hope it's at least plausible that our numbers are pretty honest, even if our design process occasionally results in sloppy graphs.
Regarding the DoW deal, I don't want to comment too publicly. I also can't say anything with confidence, as I wasn't part of the deal in any way shape or form. My perception from what I have read and heard is that both Anthropic and OpenAI have good intentions, both have loosened their prior policies over time to allow usage by the US military, and both have red lines to prohibit abuse by the US military. One place they differ is in the mechanisms employed to enforce those red lines (e.g. usage policies vs refusals vs human oversight). Each company asserts their methods are stronger than the other's, so I think we have to make our own judgments there. Accounts from the parties involved in the negotiations also conflict, so I don't think anyone's account can be trusted 100%. With that caveat, I thought this article on the DoW's POV was interesting (seems to support the notion that the breakdown wasn't over differing red lines, especially since they almost managed to salvage the deal): https://www.piratewires.com/p/inside-pentagon-anthropic-deal...
Lastly, I hope it's obvious to everyone that Anthropic is not at all a supply chain risk and the threats there were incredibly disappointing. I support them 100% and I'm glad to see them unhurt by the empty threats."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Exploiting the most prominent AI agent benchmarks"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://rdi.berkeley.edu/blog/trustworthy-benchmarks-cont/"}},"_tags":["comment","author_tedsanders","story_47733217"],"author":"tedsanders","children":[47744295,47744746],"comment_text":"> I remember the gpt-5 benchmarks and how wildly inaccurate they were data-wise. Linking one[0] that I found so that other people can remember what I am talking about. I remember some data being completely misleading or some reaching more than 100% (iirc)
Yeah, I found that slide very embarrassing. It wasn't intentionally inaccurate or misleading - just a design error made right before we went live. All the numbers on that slide were correct, and there was no problem in terms of research accuracy or data handling or reward hacking. A single bar height had the wrong value, set to its neighbor. Back then, we in the research team would generate data and graphs, and then hand them off to a separate design team, who remade the graphs in our brand style. After the GPT-5 launch with multiple embarrassingly bad graphs, I wrote an internal library so that researchers could generate graphs in our brand style directly, without the handoff. Since then our graphs have been much better.
I don't think it's unfair to assume our sloppiness in graphs translates to sloppiness in eval results. But they are different groups of people working on different timelines, so I hope it's at least plausible that our numbers are pretty honest, even if our design process occasionally results in sloppy graphs.
Regarding the DoW deal, I don't want to comment too publicly. I also can't say anything with confidence, as I wasn't part of the deal in any way shape or form. My perception from what I have read and heard is that both Anthropic and OpenAI have good intentions, both have loosened their prior policies over time to allow usage by the US military, and both have red lines to prohibit abuse by the US military. One place they differ is in the mechanisms employed to enforce those red lines (e.g. usage policies vs refusals vs human oversight). Each company asserts their methods are stronger than the other's, so I think we have to make our own judgments there. Accounts from the parties involved in the negotiations also conflict, so I don't think anyone's account can be trusted 100%. With that caveat, I thought this article on the DoW's POV was interesting (seems to support the notion that the breakdown wasn't over differing red lines, especially since they almost managed to salvage the deal): https://www.piratewires.com/p/inside-pentagon-anthropic-deal...
Lastly, I hope it's obvious to everyone that Anthropic is not at all a supply chain risk and the threats there were incredibly disappointing. I support them 100% and I'm glad to see them unhurt by the empty threats.","created_at":"2026-04-12T19:28:43Z","created_at_i":1776022123,"objectID":"47743444","parent_id":47737862,"story_id":47733217,"story_title":"Exploiting the most prominent AI agent benchmarks","story_url":"https://rdi.berkeley.edu/blog/trustworthy-benchmarks-cont/","updated_at":"2026-04-16T16:21:01Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dvt"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"I don't want to respond to 100 comments about the same thing, and this one happens to be on top, so, in my humble opinion:
(1): You don't have to be an Ed Zitron disciple to infer that OpenAI and Anthropic are likely overvalued and that Nvidia is selling everyone shovels in a gold rush. AI is a game-changing technology, but a shitty chat interface does not a company make. OpenAI and Anthropic need to recoup astronomical costs used in training these models. Models that are now being distilled[1] and are quickly becoming commoditized. (And frankly, models that were trained by torrenting copyrighted data[2], anyway.) Many have been calling this out for years: the model cannot be your product. And to be clear, OpenAI/Anthropic most definitely know this: that's why they've been aquihiring like crazy, trying to find that one team that will make the thing.
(2): Token prices are significantly subsidized and anyone that does any serious work with AI can tell you this. Go use an almost-SOTA model (a big Deepseek or Qwen model) offered by many bare-metal providers and you'll see what "true" token prices should look like. The end-state here is likely some models running locally and some running in the cloud. But the current state of OpenClaw token-vomit on top of Claude is fiscally untenable (in fact, this is why Anthropic shut it down).
(3): This is typical Dropbox HN snark[3], of which I am also often guilty of. I really don't think AI coding is a killer product and this seems very myopic\u2014engineers are an extreme minority. Imo, the closest we've seen to something revolutionary is OpenClaw, but it's janky, hard to set up, full of vulnerabilities, and you need to buy a separate computer. But there's certainly a spark there. (And that's personally the vertical I'm focusing on.)
[1] https://www.anthropic.com/news/detecting-and-preventing-dist...
[2] https://media.npr.org/assets/artslife/arts/2025/complaint.pd...
[3] https://news.ycombinator.com/item?id=9224"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"GLM-5.1: Towards Long-Horizon Tasks"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://z.ai/blog/glm-5.1"}},"_tags":["comment","author_dvt","story_47677853"],"author":"dvt","children":[47686784,47688870],"comment_text":"I don't want to respond to 100 comments about the same thing, and this one happens to be on top, so, in my humble opinion:
(1): You don't have to be an Ed Zitron disciple to infer that OpenAI and Anthropic are likely overvalued and that Nvidia is selling everyone shovels in a gold rush. AI is a game-changing technology, but a shitty chat interface does not a company make. OpenAI and Anthropic need to recoup astronomical costs used in training these models. Models that are now being distilled[1] and are quickly becoming commoditized. (And frankly, models that were trained by torrenting copyrighted data[2], anyway.) Many have been calling this out for years: the model cannot be your product. And to be clear, OpenAI/Anthropic most definitely know this: that's why they've been aquihiring like crazy, trying to find that one team that will make the thing.
(2): Token prices are significantly subsidized and anyone that does any serious work with AI can tell you this. Go use an almost-SOTA model (a big Deepseek or Qwen model) offered by many bare-metal providers and you'll see what "true" token prices should look like. The end-state here is likely some models running locally and some running in the cloud. But the current state of OpenClaw token-vomit on top of Claude is fiscally untenable (in fact, this is why Anthropic shut it down).
(3): This is typical Dropbox HN snark[3], of which I am also often guilty of. I really don't think AI coding is a killer product and this seems very myopic\u2014engineers are an extreme minority. Imo, the closest we've seen to something revolutionary is OpenClaw, but it's janky, hard to set up, full of vulnerabilities, and you need to buy a separate computer. But there's certainly a spark there. (And that's personally the vertical I'm focusing on.)
[1] https://www.anthropic.com/news/detecting-and-preventing-dist...
[2] https://media.npr.org/assets/artslife/arts/2025/complaint.pd...
[3] https://news.ycombinator.com/item?id=9224","created_at":"2026-04-08T03:44:29Z","created_at_i":1775619869,"objectID":"47684887","parent_id":47684414,"story_id":47677853,"story_title":"GLM-5.1: Towards Long-Horizon Tasks","story_url":"https://z.ai/blog/glm-5.1","updated_at":"2026-04-10T11:51:06Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"moomooskycow"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"Seems like there are three things you want here:
1. Internal agent\n2. Product features\n3. Customer-facing agent
This could all be served by one agent, but might not be as effective as having one focused agent for each need. It's not hard to build shared context for each agent to use, and it's (generally) easier to configure somewhat domain-specific agents than one superagent.
Regarding hosting your own company agent: what's the impetus for self-hosting? If it's a data privacy thing (i.e. you don't want to send data to Anthropic or other model providers), you're going to be somewhat limited. It will cost time and money to set up infrastructure for hosting your own copy of a model, and even if you use SOTA open source models like Kimi K2.5 or Deepseek V3.2 they simply aren't as capable as the SOTA models from the frontier labs.
If hosting your own company agent just means a long-running Claude Code instance on an internal server somewhere that employees and company applications can send requests to, then that's way easier. You might even want to set up an OpenClaw agent (depending on your risk appetite, they're very capable but can be a bit of a security nightmare without proper configuration and management).
SOTA right now is the Claude Agent SDK. Package it up with some company-specific skills that capture and document company idiosyncrasies, what APIs exist and how to use them, hardcoded scripts the agent can invoke for happy path tasks and operations. Compress an index of the skills into your CLAUDE.md file to increase the rate of skill invocation, and build workflows that explicitly invoke appropriate skills as often as possible.
I'm rambling. To wrap up:
1. Internal agent: a hardened OpenClaw agent that employees can chat with, or a Claude Agent SDK integration into Slack is probably best.
2. Product features: "summarize this with AI" is more of a standard LLM request pipeline than an agentic system. Lots of AI enhanced features are just one or more prompts chained together, you can do quite a bit with that and some cleaned-up well-placed corporate context.
3. Customer-facing agent: Claude Agent SDK is, again, probably the best bet if you can swing it.
Again, if you want to avoid sending data to model providers you're gonna have a good amount of infrastructure work to do. I think the OpenAI Agents SDK is more model agnostic, LangGraph too."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Best option for hosted agent in 2026?"}},"_tags":["comment","author_moomooskycow","story_46917293"],"author":"moomooskycow","comment_text":"Seems like there are three things you want here:
1. Internal agent\n2. Product features\n3. Customer-facing agent
This could all be served by one agent, but might not be as effective as having one focused agent for each need. It's not hard to build shared context for each agent to use, and it's (generally) easier to configure somewhat domain-specific agents than one superagent.
Regarding hosting your own company agent: what's the impetus for self-hosting? If it's a data privacy thing (i.e. you don't want to send data to Anthropic or other model providers), you're going to be somewhat limited. It will cost time and money to set up infrastructure for hosting your own copy of a model, and even if you use SOTA open source models like Kimi K2.5 or Deepseek V3.2 they simply aren't as capable as the SOTA models from the frontier labs.
If hosting your own company agent just means a long-running Claude Code instance on an internal server somewhere that employees and company applications can send requests to, then that's way easier. You might even want to set up an OpenClaw agent (depending on your risk appetite, they're very capable but can be a bit of a security nightmare without proper configuration and management).
SOTA right now is the Claude Agent SDK. Package it up with some company-specific skills that capture and document company idiosyncrasies, what APIs exist and how to use them, hardcoded scripts the agent can invoke for happy path tasks and operations. Compress an index of the skills into your CLAUDE.md file to increase the rate of skill invocation, and build workflows that explicitly invoke appropriate skills as often as possible.
I'm rambling. To wrap up:
1. Internal agent: a hardened OpenClaw agent that employees can chat with, or a Claude Agent SDK integration into Slack is probably best.
2. Product features: "summarize this with AI" is more of a standard LLM request pipeline than an agentic system. Lots of AI enhanced features are just one or more prompts chained together, you can do quite a bit with that and some cleaned-up well-placed corporate context.
3. Customer-facing agent: Claude Agent SDK is, again, probably the best bet if you can swing it.
Again, if you want to avoid sending data to model providers you're gonna have a good amount of infrastructure work to do. I think the OpenAI Agents SDK is more model agnostic, LangGraph too.","created_at":"2026-02-06T20:41:05Z","created_at_i":1770410465,"objectID":"46917848","parent_id":46917293,"story_id":46917293,"story_title":"Ask HN: Best option for hosted agent in 2026?","updated_at":"2026-03-05T23:31:30Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nonrandomstring"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"> "Big Tech vs. Big Copyright"
Indeed. But when do those intersect or diverge?
I don't blame him. What would you do? If I had a near perfect data\ntraining set of all the most useful books and a hungry AI to train, it\nwould be the logical step.
The reason this is news is because of the stinking hypocrisy of it\nall. It's really the same topic as the Swartz-Altman discussion here\n[0], in that these giant companies want to have it both ways.
Where is Zuckerberg's shout-out for Alexandra Elbakyan? [1] Or for\nBrewster Kahle? Or any of the wast army of people who preserve and\ncurate the vital culture of humanity by protecting it from\nintellectual property dungeons?
The colossal hypocrisy is that a company like Meta wishes to live\nunder the protective umbrella of "Intellectual Property". It wants to\nstop me just stealing it's stuff and setting up a better Facebook
Were it exposed to the same rules it wishes to live by, it would be\ntorn apart by vibrant and deserving competition within days.
All the Zuckerberg, Meta or OpenAI are doing is setting the ground for\nthe abolition of intellectual property. They are literally the\nproverbial people who will buy the rope with which to hang themselves.
(Edit. that doesn't make sense insert <proverb about buying ropes\nthat actually makes sense>)
[0] https://news.ycombinator.com/item?id=42671427
[1] https://en.wikipedia.org/wiki/Alexandra_Elbakyan"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Zuckerberg approved training Llama on LibGen [pdf]"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.377.0_1.pdf"}},"_tags":["comment","author_nonrandomstring","story_42673628"],"author":"nonrandomstring","comment_text":"> "Big Tech vs. Big Copyright"
Indeed. But when do those intersect or diverge?
I don't blame him. What would you do? If I had a near perfect data\ntraining set of all the most useful books and a hungry AI to train, it\nwould be the logical step.
The reason this is news is because of the stinking hypocrisy of it\nall. It's really the same topic as the Swartz-Altman discussion here\n[0], in that these giant companies want to have it both ways.
Where is Zuckerberg's shout-out for Alexandra Elbakyan? [1] Or for\nBrewster Kahle? Or any of the wast army of people who preserve and\ncurate the vital culture of humanity by protecting it from\nintellectual property dungeons?
The colossal hypocrisy is that a company like Meta wishes to live\nunder the protective umbrella of "Intellectual Property". It wants to\nstop me just stealing it's stuff and setting up a better Facebook
Were it exposed to the same rules it wishes to live by, it would be\ntorn apart by vibrant and deserving competition within days.
All the Zuckerberg, Meta or OpenAI are doing is setting the ground for\nthe abolition of intellectual property. They are literally the\nproverbial people who will buy the rope with which to hang themselves.
(Edit. that doesn't make sense insert <proverb about buying ropes\nthat actually makes sense>)
[0] https://news.ycombinator.com/item?id=42671427
[1] https://en.wikipedia.org/wiki/Alexandra_Elbakyan","created_at":"2025-01-12T14:56:13Z","created_at_i":1736693773,"objectID":"42673973","parent_id":42673837,"story_id":42673628,"story_title":"Zuckerberg approved training Llama on LibGen [pdf]","story_url":"https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.377.0_1.pdf","updated_at":"2026-08-17T04:07:31Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jeroenhd"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["dont","want","company","data","sent","to","openai"],"value":"I don't think it matters how fancy you make your LLM, to be honest.
If you turn an LLM into a person, you may have a ethical and legal basis for treating that LLM like a person. There's no law about artificial intelligence being sentient or not, but law applies only to people, so that'd be the supreme court case of the century. I remember the Star Trek TNG episode about this topic and while the answer was perhaps more obvious with mister Data, the best arguments for and against synthetic consciousnesses have all been made in that episode.
The law doesn't care for how human-like programmers may think their program is, and neither should it in my opinion. What matters to the law is that the output is a result of an automated process, which comes with a completely separate set of rules and conditions compared to fair use.
The complexity of the program isn't a very good legal defence in my opinion because there's no clear line when the complexity would be enough to be considered human like. You could, for example, also claim that a computer is just very good at doing imitations, just like a person can be good at doing imitations on stage, and that an mp3 file is just an elaborate imitation act.
Even with a digital system identical to a physical system I don't think you can state human-ness as an argument. A tape recorder is just a sophisticated, automated way of sending an electric field through a magnet, similar to what a human can do with a dynamo and a spool of tape; a sort of delayed-action theremin, which would turn it into a musical signal. In turn, a neural network can be solved by human brain power if you pay enough people to work on a single iteration for an entire year. Almost everything a computer can do is just a sophisticated way of doing what humans are already doing, so I don't see why this is different when it comes to this topic.
There's no obvious "this is human like now" threshold and I doubt there will be until we know exactly how the human brain works.
When it comes to AI generated works, we don't currently know where the line between copyright violation and copyright exemption lies. If this goes through, it's the third major lawsuit of its type, the other two being actions against Stable Diffusion by artists to prevent it from producing derived works from their art.
IANA but I think you can assume that the "but computers are just like digital humans" approach won't fly in court. That's not really important, though; both sides of the coin are already having clever lawyers write up legal defences for their points of view, and it's more than likely that some other factors ("is a model a derived work" (probably) and "is the output of a model a derived work" (who knows!)) will decide the future of AI and copyright. The interesting thing is that academic research is essentially exempt from copyright law, so nobody can demand takedown of their content from research data sets, but whether the commercial branch of AI companies can use their academically generated models to serve their customers?
As an upside, I don't think the death of ChatGPT is the end of AI. This whole scenario could've easily been avoided if AI companies paid for their data set or had restricted themselves to works they had the license to (public domain, CC0, etc.), and OpenAI in particular has been pretty brazen in their "we'll see about it if it ever comes up" approach. Companies like Github are probably in the best place legally, where their users are already signing off on "we can take your content and do whatever the fuck we want" terms and conditions."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"New York Times considers legal action against OpenAI as copyright tensions swirl"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["openai"],"value":"https://www.npr.org/2023/08/16/1194202562/new-york-times-considers-legal-action-against-openai-as-copyright-tensions-swirl"}},"_tags":["comment","author_jeroenhd","story_37154840"],"author":"jeroenhd","comment_text":"I don't think it matters how fancy you make your LLM, to be honest.
If you turn an LLM into a person, you may have a ethical and legal basis for treating that LLM like a person. There's no law about artificial intelligence being sentient or not, but law applies only to people, so that'd be the supreme court case of the century. I remember the Star Trek TNG episode about this topic and while the answer was perhaps more obvious with mister Data, the best arguments for and against synthetic consciousnesses have all been made in that episode.
The law doesn't care for how human-like programmers may think their program is, and neither should it in my opinion. What matters to the law is that the output is a result of an automated process, which comes with a completely separate set of rules and conditions compared to fair use.
The complexity of the program isn't a very good legal defence in my opinion because there's no clear line when the complexity would be enough to be considered human like. You could, for example, also claim that a computer is just very good at doing imitations, just like a person can be good at doing imitations on stage, and that an mp3 file is just an elaborate imitation act.
Even with a digital system identical to a physical system I don't think you can state human-ness as an argument. A tape recorder is just a sophisticated, automated way of sending an electric field through a magnet, similar to what a human can do with a dynamo and a spool of tape; a sort of delayed-action theremin, which would turn it into a musical signal. In turn, a neural network can be solved by human brain power if you pay enough people to work on a single iteration for an entire year. Almost everything a computer can do is just a sophisticated way of doing what humans are already doing, so I don't see why this is different when it comes to this topic.
There's no obvious "this is human like now" threshold and I doubt there will be until we know exactly how the human brain works.
When it comes to AI generated works, we don't currently know where the line between copyright violation and copyright exemption lies. If this goes through, it's the third major lawsuit of its type, the other two being actions against Stable Diffusion by artists to prevent it from producing derived works from their art.
IANA but I think you can assume that the "but computers are just like digital humans" approach won't fly in court. That's not really important, though; both sides of the coin are already having clever lawyers write up legal defences for their points of view, and it's more than likely that some other factors ("is a model a derived work" (probably) and "is the output of a model a derived work" (who knows!)) will decide the future of AI and copyright. The interesting thing is that academic research is essentially exempt from copyright law, so nobody can demand takedown of their content from research data sets, but whether the commercial branch of AI companies can use their academically generated models to serve their customers?
As an upside, I don't think the death of ChatGPT is the end of AI. This whole scenario could've easily been avoided if AI companies paid for their data set or had restricted themselves to works they had the license to (public domain, CC0, etc.), and OpenAI in particular has been pretty brazen in their "we'll see about it if it ever comes up" approach. Companies like Github are probably in the best place legally, where their users are already signing off on "we can take your content and do whatever the fuck we want" terms and conditions.","created_at":"2023-08-17T15:34:41Z","created_at_i":1692286481,"objectID":"37163279","parent_id":37159491,"story_id":37154840,"story_title":"New York Times considers legal action against OpenAI as copyright tensions swirl","story_url":"https://www.npr.org/2023/08/16/1194202562/new-york-times-considers-legal-action-against-openai-as-copyright-tensions-swirl","updated_at":"2024-09-20T14:51:54Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"marioskales"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["sent","to"],"value":"Hey HN,
I'm Mario from Vienna. I\u2019m the designated "IT guy" for my family (the guy they call to plug in a Scart cable or to setup Chrome).
A few weeks ago, I spent over two hours wrestling with Docker (and some for my at this time unknown Terminal commands) just to get a CLI-based AI agent running. It ate all my RAM, and I had a realization: my wife, my non-technical clients, or my 6-year-old son could never set this up. If I deployed something like this, I'd spend my life doing setup-support (and this isn't just about finding the HDMI 1 to HDMI 2 input on the remote control).
I\u2019ve been "vibecoding" with LLMs for a while the last 2 years, mostly building common landingpages or simple retro games to show my son that we live in a wild time where you can build a lot of 'entertaining' stuff just by prompting. I decided to pivot an old, failed Laravel SaaS of mine (Skales) into a native desktop app to solve this setup hell.
The only goal I wanted to reach: an autonomous AI agent that installs like a normal app (.exe / .dmg). No PhD required. Download, Install, Done.
It actually worked. My 60+ year old mom got it running instantly (so.. is Skales now a Grandma-Approved AI Agent?), and my 6yr used the built-in Coding Skill to make his own retro game (I wouldn't say its really a 'game' when it's basically a one level copy of Super Mario - but he loves it).
What it does:
ReAct autopilot, bi-temporal memory, browser automation (Playwright), and native integrations (Gmail, Telegram, Calendar).
BYOK: Works with OpenRouter, OpenAI, Anthropic, or local Ollama.
Tech stack: Electron + Next.js + Node.js (I managed to keep idle RAM around ~300MB).
Everything is stored locally in ~/.skales-data.
I know that being a "design guy who vibecoded an Electron app" (or Marketing-Dude) is basically fresh meat for the HN crowd. Feel free to roast the architecture! But my main goal was just making these incredible AI tools accessible for everyday tasks (formatting resumes, sending PDFs, building simple games) without touching a terminal or moving from one tool to another tool.
License note: The project is BSL-1.1 (Source-available, free for personal use). I chose this simply because I don't want a giant company to just take the repo, build their brand around Skales and commercially resell it, but I want the code out there for the community to use and learn from.
(excuse my English - i'm not a native speaker)
Would love your honest feedback on the UX (could be maybe better?)!
GitHub: https://github.com/skalesapp/skales\nWebsite: https://skales.app"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Skales \u2013 Local AI agent desktop app (.exe/.dmg, 300MB idle RAM)"}},"_tags":["story","author_marioskales","story_47299856","show_hn"],"author":"marioskales","children":[47300766,47302104,47303297,47320882,47330744,47333320,47384911],"created_at":"2026-03-08T18:45:01Z","created_at_i":1772995501,"num_comments":6,"objectID":"47299856","points":2,"story_id":47299856,"story_text":"Hey HN,
I'm Mario from Vienna. I\u2019m the designated "IT guy" for my family (the guy they call to plug in a Scart cable or to setup Chrome).
A few weeks ago, I spent over two hours wrestling with Docker (and some for my at this time unknown Terminal commands) just to get a CLI-based AI agent running. It ate all my RAM, and I had a realization: my wife, my non-technical clients, or my 6-year-old son could never set this up. If I deployed something like this, I'd spend my life doing setup-support (and this isn't just about finding the HDMI 1 to HDMI 2 input on the remote control).
I\u2019ve been "vibecoding" with LLMs for a while the last 2 years, mostly building common landingpages or simple retro games to show my son that we live in a wild time where you can build a lot of 'entertaining' stuff just by prompting. I decided to pivot an old, failed Laravel SaaS of mine (Skales) into a native desktop app to solve this setup hell.
The only goal I wanted to reach: an autonomous AI agent that installs like a normal app (.exe / .dmg). No PhD required. Download, Install, Done.
It actually worked. My 60+ year old mom got it running instantly (so.. is Skales now a Grandma-Approved AI Agent?), and my 6yr used the built-in Coding Skill to make his own retro game (I wouldn't say its really a 'game' when it's basically a one level copy of Super Mario - but he loves it).
What it does:
ReAct autopilot, bi-temporal memory, browser automation (Playwright), and native integrations (Gmail, Telegram, Calendar).
BYOK: Works with OpenRouter, OpenAI, Anthropic, or local Ollama.
Tech stack: Electron + Next.js + Node.js (I managed to keep idle RAM around ~300MB).
Everything is stored locally in ~/.skales-data.
I know that being a "design guy who vibecoded an Electron app" (or Marketing-Dude) is basically fresh meat for the HN crowd. Feel free to roast the architecture! But my main goal was just making these incredible AI tools accessible for everyday tasks (formatting resumes, sending PDFs, building simple games) without touching a terminal or moving from one tool to another tool.
License note: The project is BSL-1.1 (Source-available, free for personal use). I chose this simply because I don't want a giant company to just take the repo, build their brand around Skales and commercially resell it, but I want the code out there for the community to use and learn from.
(excuse my English - i'm not a native speaker)
Would love your honest feedback on the UX (could be maybe better?)!
GitHub: https://github.com/skalesapp/skales\nWebsite: https://skales.app","title":"Show HN: Skales \u2013 Local AI agent desktop app (.exe/.dmg, 300MB idle RAM)","updated_at":"2026-05-14T17:52:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"blagie"},"comment_text":{"matchLevel":"none","matchedWords":[],"value":"As much as I agree with the conclusion (my AGPL code was stolen by OpenAI, as was a lot of other code), your logic is extreme. This is not good. A more reasoned argument is more convincing.
1. "Fair use"
It is fair use to use snippets of my code in other projects. AGPL doesn't change that. If I have a 3-line sort routine, and Copilot copies that verbatim, it's okay.
The place and reason Copilot is not okay is because I can ask it to spit out code which, in some versions, was almost verbatim code I wrote, and in others, a paraphrase. The same abstractions and high-level design I invented is copied by GPT.
It's like writing a book set in the Star Trek universe and (not as a parody), changing "Spock" to "Spork" and "Kirk" to "Dirk."
It's a problem for me because I compete with commercial systems, and chose the AGPL for very specific reasons: People have a right to know how their data is analyzed, and AGPL preserves that. Copilot allows companies to quickly replicate and whitewash my code.
If Microsoft uses a snippet in Windows -- e.g. a small use in an unrelated context, fair use clearly allows that. As much as I don't want my code stolen, degrading fair use would be the far greater evil.
2. "The reason why OpenAI has a non-profit part is to allow them to claim that they do non-profit research, which fair-use doctrine requires to work."
No, OpenAI was set up as a genuine, if ill-formed, non-profit. When it clearly became big $$$, bad people (likely breaking the law) formed a for-profit arm to personally profit from it. The non-profit arm is there for historical reasons.
Your statement is false, and there is ample historical record that what you're stating is not how it happened.
That's dangerous, since the wrong lessons would be drawn. The lesson this reiterates, which I learned many times the hard way, is that for non-profits, checks-and-balances matter. If you don't want your non-profit going south, this means everything -- licenses, corporate charter, bylaws, contracts, etc. -- need to be properly structured to act as a permanent network of checks-and-balances so the organization cannot go south even if bad people are involved, and that incentives are structured to support good behavior.
We do need a for-profit like the original OpenAI, but properly structured and resourced."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"French court issues damages award for violation of GPL"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://heathermeeker.com/2024/02/17/french-court-issues-damages-award-for-violation-of-gpl/"}},"_tags":["comment","author_blagie","story_39587344"],"author":"blagie","children":[39590028],"comment_text":"As much as I agree with the conclusion (my AGPL code was stolen by OpenAI, as was a lot of other code), your logic is extreme. This is not good. A more reasoned argument is more convincing.
1. "Fair use"
It is fair use to use snippets of my code in other projects. AGPL doesn't change that. If I have a 3-line sort routine, and Copilot copies that verbatim, it's okay.
The place and reason Copilot is not okay is because I can ask it to spit out code which, in some versions, was almost verbatim code I wrote, and in others, a paraphrase. The same abstractions and high-level design I invented is copied by GPT.
It's like writing a book set in the Star Trek universe and (not as a parody), changing "Spock" to "Spork" and "Kirk" to "Dirk."
It's a problem for me because I compete with commercial systems, and chose the AGPL for very specific reasons: People have a right to know how their data is analyzed, and AGPL preserves that. Copilot allows companies to quickly replicate and whitewash my code.
If Microsoft uses a snippet in Windows -- e.g. a small use in an unrelated context, fair use clearly allows that. As much as I don't want my code stolen, degrading fair use would be the far greater evil.
2. "The reason why OpenAI has a non-profit part is to allow them to claim that they do non-profit research, which fair-use doctrine requires to work."
No, OpenAI was set up as a genuine, if ill-formed, non-profit. When it clearly became big $$$, bad people (likely breaking the law) formed a for-profit arm to personally profit from it. The non-profit arm is there for historical reasons.
Your statement is false, and there is ample historical record that what you're stating is not how it happened.
That's dangerous, since the wrong lessons would be drawn. The lesson this reiterates, which I learned many times the hard way, is that for non-profits, checks-and-balances matter. If you don't want your non-profit going south, this means everything -- licenses, corporate charter, bylaws, contracts, etc. -- need to be properly structured to act as a permanent network of checks-and-balances so the organization cannot go south even if bad people are involved, and that incentives are structured to support good behavior.
We do need a for-profit like the original OpenAI, but properly structured and resourced.","created_at":"2024-03-04T12:38:27Z","created_at_i":1709555907,"objectID":"39589709","parent_id":39589197,"story_id":39587344,"story_title":"French court issues damages award for violation of GPL","story_url":"https://heathermeeker.com/2024/02/17/french-court-issues-damages-award-for-violation-of-gpl/","updated_at":"2024-09-20T16:34:23Z"}],"hitsPerPage":20,"nbHits":114,"nbPages":6,"page":0,"params":"query=dont+want+company+data+sent+to+OpenAI&advancedSyntax=true&analyticsTags=backend","processingTimeMS":76,"processingTimingsMS":{"_request":{"roundTrip":16},"afterFetch":{"format":{"highlighting":4,"total":4},"merge":{"mergeLoop":{"prepareNextHit":4,"total":4},"total":6},"total":6},"fetch":{"query":17,"scanning":51,"total":69},"total":76},"query":"dont want company data sent to OpenAI","serverTimeMS":81}