I asked Google if MA does LLM tutoring and got back this answer:
> Math Academy does not offer Large Language Model (LLM) tutoring. While the company advertises itself as "AI-powered," this is in reference to a machine-learning-based adaptive learning system, not an interactive LLM tutor.
And here is a HN comment that indicates LLMs are a complement to MA, not part of it: https://news.ycombinator.com/item?id=43281240"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["llm"],"value":"LLMs are the ultimate demoware"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["llm"],"value":"https://blog.charliemeyer.co/llms-are-the-ultimate-demoware/"}},"_tags":["comment","author_getnormality","story_45437113"],"author":"getnormality","children":[45438922],"comment_text":"Everything described there sounds like old-school adaptive algorithms. I don't see anything about generative AI or LLMs.
I asked Google if MA does LLM tutoring and got back this answer:
> Math Academy does not offer Large Language Model (LLM) tutoring. While the company advertises itself as "AI-powered," this is in reference to a machine-learning-based adaptive learning system, not an interactive LLM tutor.
And here is a HN comment that indicates LLMs are a complement to MA, not part of it: https://news.ycombinator.com/item?id=43281240","created_at":"2025-10-01T15:26:57Z","created_at_i":1759332417,"objectID":"45438868","parent_id":45438791,"story_id":45437113,"story_title":"LLMs are the ultimate demoware","story_url":"https://blog.charliemeyer.co/llms-are-the-ultimate-demoware/","updated_at":"2026-03-05T22:47:55Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"alexsouthmayd"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"Hi HN, I\u2019m Alex Southmayd, the founder of Bloomy (https://bloomylearning.com) \u2013 an AI-powered mastery-learning platform for K-12 students. Bloomy provides students with an AI tutor alongside adaptive curriculum (right now Math, English Language Arts, and Writing).
How it works: we diagnose students\u2019 skill gaps, place them on personalized learning paths, and give them standards-aligned lessons and a Socratic AI tutor that scaffolds their learning without just giving away the answer.
The goal is to solve the Bloom 2-sigma problem (https://en.wikipedia.org/wiki/Bloom%27s_2_sigma_problem) with AI.
Short launch video: https://tinyurl.com/bloomylearning
Longer product demo: https://youtu.be/XHvoKt6qMeo
Families access for Bloomy: https://bloomylearning.com/families
I started as a teacher. I taught 7th-grade English and writing with Teach For America, and every day I struggled to deliver differentiated instruction to 30 students with 30 different sets of needs. Some students needed remediation, some needed acceleration, and many needed a tutor sitting next to them helping them reason through the next step.\nBenjamin Bloom\u2019s two-sigma result\u2014that one-on-one tutoring can produce much better outcomes than conventional classroom instruction\u2014always felt intuitively true to me. The hard part was making that kind of attention affordable and available to every child.
Then AI changed the cost curve. When I saw schools such as Alpha organize academics around mastery rather than seat time, the model clicked. If you\u2019ve heard of Alpha School, that is directionally the kind of learning model that inspired us. But I kept thinking about the families and schools that already exist: homeschool families, microschools, hybrid schools, and regular classrooms where most children are today.
Most students and teachers see learning gaps at the wrong resolution. They get a grade, percentile, benchmark score, or broad standard\u2014not \u201cthis is the next skill this student should learn.\u201d Existing personalized-learning products often feel like digital worksheets: they provide plenty of practice, but not much diagnosis or teaching. Very few have AI tutors providing the core instruction.\nBloomy starts with a diagnostic\u2014we integrate with third-party assessments and provide our own\u2014and creates a learning path for each student. Students work one skill at a time, receive a short lesson, practice at an adaptive difficulty, and only move forward after demonstrating at least 90% mastery. The learning path updates as the student works, based on their performance and our knowledge graph of skill prerequisites (built in collaboration with Learning Commons / Chan Zuckerberg Initiative).
Each skill has three stages. Base Camp teaches the concept with worked examples. Climb provides guided practice and Socratic support. Summit is an independent ten-question mastery assessment with no hints or AI assistance. Students need to achieve 90% on the Summit to advance. If they struggle too much, they\u2019ll be routed to a different skill better suited for their level.
BloomyBot is not a blank chat window but rather a live, interactive, and observant digital tutor. During practice, it receives the active passage or problem, the question, the student\u2019s attempt, an authored explanation, and relevant misconception context. It follows a scaffolded tutoring ladder: first asking what the student tried, then pointing toward the concept, suggesting a strategy, working through one step together, and only providing heavier scaffolding after the student has struggled, adapting to and learning from the student along the way. Students can interrupt it, and we\u2019ve begun to roll out multilingual support for Spanish, French, and a few other more niche languages that customers have asked for.
We currently use a variety of Anthropic and OpenAI models for BloomyBot. The tutor is restricted to the current lesson, redirects unrelated questions, limits conversation length, and is unavailable during mastery assessments. The language model does not choose the curriculum or decide whether a student has mastered a skill.
That separation is important. A conventionally \u201chelpful\u201d AI response can be a bad tutoring response: if it gives away the answer, the student completes the task but may not learn anything. Our goal is not to build a homework-answering chatbot. It is to put AI inside a structured loop of diagnosis, instruction, practice, feedback, and independent mastery.
LLMs can still be wrong, and we do not claim our constraints eliminate that. We reduce the surface area by grounding BloomyBot in authored lesson content, keeping it on topic, logging conversations, and removing it from assessments. Teachers and parents can review tutoring activity, students can report problems, and safety signals trigger human alerts and a backup audit.\nWe also do not see Bloomy as a replacement for teachers, parents, or human tutors. A good human tutor is better. The narrower question we are testing is whether, during a bounded learning session a student would already be doing, a context-aware tutor can provide better help than static \u201ccorrect/incorrect\u201d feedback. Longer term, the question becomes more whether a student would perform better with one-on-one AI tutoring (at least in certain aspects of the curriculum) than with many-to-one instruction in a medium- or large-sized classroom.
Bloomy is now being used across several settings: traditional districts, charter schools, hybrid schools, microschools, homeschools, and families looking for additional academic support. In an early pilot at a charter school in Massachusetts serving ~150 students in grades 6 through 8, students averaged roughly 1.8 times the expected winter-to-spring NWEA MAP growth. This was an observational pilot, not a randomized study, so we treat it as an encouraging signal rather than proof that Bloomy caused the difference.
Parents and teachers can see what a student has mastered, what is in progress, and where support may be needed. We have found that adults generally do not want another generic score; they want to know which small number of skills deserve attention this week.
Bloomy makes money through family subscriptions and school licensing. ELA costs $39/month or $279/year per learner, and Writing Studio costs $19/month or $139/year. Math is scheduled to launch July 31 at the same price as ELA. Schools and microschools pay per student, with pricing varying by subject coverage, enrollment, rostering, and implementation needs.
Because children use Bloomy, we collect learning responses, progress data, and tutoring conversations. We do not sell personal information, use child data for behavioral advertising, or permit model providers to train general-purpose models on identifiable child data sent by Bloomy. We have Zero Data Retention agreements with both Anthropic and OpenAI. Parents and schools can request access, export, correction, or deletion under the applicable account or school agreement.
More background on me: after Teach For America, I taught and designed GMAT and GRE curriculum for Manhattan Prep / Kaplan, led the driver acquisition team for Lyft\u2019s New England markets, completed an MBA at Stanford, and led AI transformation projects at McKinsey (so when models finally became good enough this past January to achieve the kinds of things I am pursuing with Bloomy, I was in the right place at the right time to begin building). Bloomy brings together the different parts of my career that I care most about: educational outcomes, learning design, building products, and getting useful technology into people\u2019s hands.
I\u2019d especially value feedback from parents, teachers, engineers working on child-facing AI, and people who have built tutoring, assessment, or adaptive-learning systems. Does the separation between guided AI help and independent mastery make sense? Where do you see the greatest potential with AI in education? Where are our safeguards insufficient? What evidence or product behavior would you need to trust something like this with a student?
Certainly there are many dangers and pitfalls we must beware of, too, but I believe we can really move the needle in K-12 (for the first time in a long time) if we use AI responsibly and intelligently."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["learning"],"value":"Launch HN: Bloomy (YC S26) \u2013 AI-powered mastery learning for K-12"}},"_tags":["story","author_alexsouthmayd","story_48981136","launch_hn"],"author":"alexsouthmayd","children":[48981267,48981405,48981468,48981492,48981705,48981843,48982554,48982555,48982734,48982894,48983239,48983323,48983512,48984520,48984624,48984769,48984947,48984967,48987201,48988338,48988345,48988375,48988397,48988633,48988864,48990006,48990385,48991004,48992902,48992912,49012793,49017646],"created_at":"2026-07-20T16:32:10Z","created_at_i":1784565130,"num_comments":107,"objectID":"48981136","points":102,"story_id":48981136,"story_text":"Hi HN, I\u2019m Alex Southmayd, the founder of Bloomy (https://bloomylearning.com) \u2013 an AI-powered mastery-learning platform for K-12 students. Bloomy provides students with an AI tutor alongside adaptive curriculum (right now Math, English Language Arts, and Writing).
How it works: we diagnose students\u2019 skill gaps, place them on personalized learning paths, and give them standards-aligned lessons and a Socratic AI tutor that scaffolds their learning without just giving away the answer.
The goal is to solve the Bloom 2-sigma problem (https://en.wikipedia.org/wiki/Bloom%27s_2_sigma_problem) with AI.
Short launch video: https://tinyurl.com/bloomylearning
Longer product demo: https://youtu.be/XHvoKt6qMeo
Families access for Bloomy: https://bloomylearning.com/families
I started as a teacher. I taught 7th-grade English and writing with Teach For America, and every day I struggled to deliver differentiated instruction to 30 students with 30 different sets of needs. Some students needed remediation, some needed acceleration, and many needed a tutor sitting next to them helping them reason through the next step.\nBenjamin Bloom\u2019s two-sigma result\u2014that one-on-one tutoring can produce much better outcomes than conventional classroom instruction\u2014always felt intuitively true to me. The hard part was making that kind of attention affordable and available to every child.
Then AI changed the cost curve. When I saw schools such as Alpha organize academics around mastery rather than seat time, the model clicked. If you\u2019ve heard of Alpha School, that is directionally the kind of learning model that inspired us. But I kept thinking about the families and schools that already exist: homeschool families, microschools, hybrid schools, and regular classrooms where most children are today.
Most students and teachers see learning gaps at the wrong resolution. They get a grade, percentile, benchmark score, or broad standard\u2014not \u201cthis is the next skill this student should learn.\u201d Existing personalized-learning products often feel like digital worksheets: they provide plenty of practice, but not much diagnosis or teaching. Very few have AI tutors providing the core instruction.\nBloomy starts with a diagnostic\u2014we integrate with third-party assessments and provide our own\u2014and creates a learning path for each student. Students work one skill at a time, receive a short lesson, practice at an adaptive difficulty, and only move forward after demonstrating at least 90% mastery. The learning path updates as the student works, based on their performance and our knowledge graph of skill prerequisites (built in collaboration with Learning Commons / Chan Zuckerberg Initiative).
Each skill has three stages. Base Camp teaches the concept with worked examples. Climb provides guided practice and Socratic support. Summit is an independent ten-question mastery assessment with no hints or AI assistance. Students need to achieve 90% on the Summit to advance. If they struggle too much, they\u2019ll be routed to a different skill better suited for their level.
BloomyBot is not a blank chat window but rather a live, interactive, and observant digital tutor. During practice, it receives the active passage or problem, the question, the student\u2019s attempt, an authored explanation, and relevant misconception context. It follows a scaffolded tutoring ladder: first asking what the student tried, then pointing toward the concept, suggesting a strategy, working through one step together, and only providing heavier scaffolding after the student has struggled, adapting to and learning from the student along the way. Students can interrupt it, and we\u2019ve begun to roll out multilingual support for Spanish, French, and a few other more niche languages that customers have asked for.
We currently use a variety of Anthropic and OpenAI models for BloomyBot. The tutor is restricted to the current lesson, redirects unrelated questions, limits conversation length, and is unavailable during mastery assessments. The language model does not choose the curriculum or decide whether a student has mastered a skill.
That separation is important. A conventionally \u201chelpful\u201d AI response can be a bad tutoring response: if it gives away the answer, the student completes the task but may not learn anything. Our goal is not to build a homework-answering chatbot. It is to put AI inside a structured loop of diagnosis, instruction, practice, feedback, and independent mastery.
LLMs can still be wrong, and we do not claim our constraints eliminate that. We reduce the surface area by grounding BloomyBot in authored lesson content, keeping it on topic, logging conversations, and removing it from assessments. Teachers and parents can review tutoring activity, students can report problems, and safety signals trigger human alerts and a backup audit.\nWe also do not see Bloomy as a replacement for teachers, parents, or human tutors. A good human tutor is better. The narrower question we are testing is whether, during a bounded learning session a student would already be doing, a context-aware tutor can provide better help than static \u201ccorrect/incorrect\u201d feedback. Longer term, the question becomes more whether a student would perform better with one-on-one AI tutoring (at least in certain aspects of the curriculum) than with many-to-one instruction in a medium- or large-sized classroom.
Bloomy is now being used across several settings: traditional districts, charter schools, hybrid schools, microschools, homeschools, and families looking for additional academic support. In an early pilot at a charter school in Massachusetts serving ~150 students in grades 6 through 8, students averaged roughly 1.8 times the expected winter-to-spring NWEA MAP growth. This was an observational pilot, not a randomized study, so we treat it as an encouraging signal rather than proof that Bloomy caused the difference.
Parents and teachers can see what a student has mastered, what is in progress, and where support may be needed. We have found that adults generally do not want another generic score; they want to know which small number of skills deserve attention this week.
Bloomy makes money through family subscriptions and school licensing. ELA costs $39/month or $279/year per learner, and Writing Studio costs $19/month or $139/year. Math is scheduled to launch July 31 at the same price as ELA. Schools and microschools pay per student, with pricing varying by subject coverage, enrollment, rostering, and implementation needs.
Because children use Bloomy, we collect learning responses, progress data, and tutoring conversations. We do not sell personal information, use child data for behavioral advertising, or permit model providers to train general-purpose models on identifiable child data sent by Bloomy. We have Zero Data Retention agreements with both Anthropic and OpenAI. Parents and schools can request access, export, correction, or deletion under the applicable account or school agreement.
More background on me: after Teach For America, I taught and designed GMAT and GRE curriculum for Manhattan Prep / Kaplan, led the driver acquisition team for Lyft\u2019s New England markets, completed an MBA at Stanford, and led AI transformation projects at McKinsey (so when models finally became good enough this past January to achieve the kinds of things I am pursuing with Bloomy, I was in the right place at the right time to begin building). Bloomy brings together the different parts of my career that I care most about: educational outcomes, learning design, building products, and getting useful technology into people\u2019s hands.
I\u2019d especially value feedback from parents, teachers, engineers working on child-facing AI, and people who have built tutoring, assessment, or adaptive-learning systems. Does the separation between guided AI help and independent mastery make sense? Where do you see the greatest potential with AI in education? Where are our safeguards insufficient? What evidence or product behavior would you need to trust something like this with a student?
Certainly there are many dangers and pitfalls we must beware of, too, but I believe we can really move the needle in K-12 (for the first time in a long time) if we use AI responsibly and intelligently.","title":"Launch HN: Bloomy (YC S26) \u2013 AI-powered mastery learning for K-12","updated_at":"2026-08-07T04:06:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mncharity"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"Years ago I was watching so-called "best practices" videos on teaching estimation in early primary. I was astonished. Often the skill being coached seemed to be participative performance in a collaborative pretense of understanding. What to do, and not do, so we can both pretend learning has occurred and move on. With the teachers themselves largely unclear on the concepts.
Ongoing formative assessment is usually closely associated with instruction, with attendant conflicts of interest. I wonder if say LLMs, might transformatively permit fine-grain and adaptive assessment. So a "what have you been learning?" dialog for objective identification, and then assessment across the objective neighborhood. With potential for tutoring of course. But here emphasizing a possibility for rapid verification of learning."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["learning"],"value":"Lots of people in education disagree with the premise of maximizing learning"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["learning"],"value":"https://www.justinmath.com/maximizing-learning-vs-other-things/"}},"_tags":["comment","author_mncharity","story_40916904"],"author":"mncharity","children":[40921804,40922012],"comment_text":"Years ago I was watching so-called "best practices" videos on teaching estimation in early primary. I was astonished. Often the skill being coached seemed to be participative performance in a collaborative pretense of understanding. What to do, and not do, so we can both pretend learning has occurred and move on. With the teachers themselves largely unclear on the concepts.
Ongoing formative assessment is usually closely associated with instruction, with attendant conflicts of interest. I wonder if say LLMs, might transformatively permit fine-grain and adaptive assessment. So a "what have you been learning?" dialog for objective identification, and then assessment across the objective neighborhood. With potential for tutoring of course. But here emphasizing a possibility for rapid verification of learning.","created_at":"2024-07-09T22:00:31Z","created_at_i":1720562431,"objectID":"40921700","parent_id":40921090,"story_id":40916904,"story_title":"Lots of people in education disagree with the premise of maximizing learning","story_url":"https://www.justinmath.com/maximizing-learning-vs-other-things/","updated_at":"2024-09-20T17:23:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"fraggler"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"The core idea is this:
Truly general intelligence requires the ability to build an unbounded number of solutions. For a finite system to achieve this, it needs a mechanism for unbounded generation, like recursion (similar to how a Turing machine operates). General intelligence also requires constant adaptation. So how do you get both? The paper proposes it arises from the dynamic interaction between an adaptive continuous substrate (like the human brain or an ANN) and an internalized symbolic framework (like human language).
The "engine" of this process, according to the ESC framework, is recursive symbolic generation. The substrate learns to:\n1. Sequentially generate symbolic sequences (like words forming thoughts or sentences).\n2. Process these sequences.\n3. Evaluate them based on internal rules and goals.
This recursive loop allows the system to effectively function as a powerful, discrete symbolic processor, capable of navigating vast combinatorial spaces and constructing structured solutions for diverse problems\u2014essentially, to think and reason in a general-purpose way.
Why this might be interesting:\n- It tries to bridge the gap between connectionist learning (like in ANNs) and symbolic competence (rule-based reasoning).\n- It offers a lens on why language seems so crucial for human thought.\n- It sheds light on the surprising abilities emerging in LLMs (which learn only from text) as a key piece of evidence.\n- It defines GI functionally, focusing on what it does (generates novel information to solve problems across unbounded domains)."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["llm"],"value":"Beyond \"stochastic parrots\": LLMs reveal language's role in general intelligence"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://osf.io/preprints/psyarxiv/86xsj_v24"}},"_tags":["comment","author_fraggler","story_44161689"],"author":"fraggler","comment_text":"The core idea is this:
Truly general intelligence requires the ability to build an unbounded number of solutions. For a finite system to achieve this, it needs a mechanism for unbounded generation, like recursion (similar to how a Turing machine operates). General intelligence also requires constant adaptation. So how do you get both? The paper proposes it arises from the dynamic interaction between an adaptive continuous substrate (like the human brain or an ANN) and an internalized symbolic framework (like human language).
The "engine" of this process, according to the ESC framework, is recursive symbolic generation. The substrate learns to:\n1. Sequentially generate symbolic sequences (like words forming thoughts or sentences).\n2. Process these sequences.\n3. Evaluate them based on internal rules and goals.
This recursive loop allows the system to effectively function as a powerful, discrete symbolic processor, capable of navigating vast combinatorial spaces and constructing structured solutions for diverse problems\u2014essentially, to think and reason in a general-purpose way.
Why this might be interesting:\n- It tries to bridge the gap between connectionist learning (like in ANNs) and symbolic competence (rule-based reasoning).\n- It offers a lens on why language seems so crucial for human thought.\n- It sheds light on the surprising abilities emerging in LLMs (which learn only from text) as a key piece of evidence.\n- It defines GI functionally, focusing on what it does (generates novel information to solve problems across unbounded domains).","created_at":"2025-06-02T21:04:39Z","created_at_i":1748898279,"objectID":"44163022","parent_id":44161689,"story_id":44161689,"story_title":"Beyond \"stochastic parrots\": LLMs reveal language's role in general intelligence","story_url":"https://osf.io/preprints/psyarxiv/86xsj_v24","updated_at":"2025-06-03T09:00:28Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Aeyxen"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"I understand your perspective as a marketer, but I think you're creating a false dichotomy. Yes, persuasion tech has stronger financial incentives, but that doesn't prevent beneficial applications from emerging simultaneously.
The "super tutor" isn't some distant fantasy - millions already use ChatGPT, Claude and similar tools daily for personalized learning. They're imperfect but genuinely helpful for programming, languages, math, and countless other topics.
Look at what happened with YouTube: millions of people transformed themselves into programmers, musicians, mechanics, and countless other professions through free video tutorials. Khan Academy revolutionized math education. Coursera and edX brought university courses to anyone with internet. This wasn't utopian thinking - it was practical technology solving real educational problems at scale.
What's different now is that LLMs enable the missing piece: personalization. The one-on-one adaptive experience that was previously limited to those who could afford human tutors at $50-100/hour is now available to anyone at negligible marginal cost.
Your skepticism about cancer applications too ignores the technological trajectory we've been on for decades. Just as YouTube and online platforms democratized education, technology has been steadily dismantling bottlenecks in medical research.
The human genome project initially cost $3 billion and took 13 years. Today you can sequence a genome for under $1,000 in days. This wasn't utopian thinking; it was technological progress following its natural course.
Think what LLMs will do here."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["llm"],"value":"LLMs are more persuasive than incentivized human persuaders"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2505.09662"}},"_tags":["comment","author_Aeyxen","story_44016621"],"author":"Aeyxen","comment_text":"I understand your perspective as a marketer, but I think you're creating a false dichotomy. Yes, persuasion tech has stronger financial incentives, but that doesn't prevent beneficial applications from emerging simultaneously.
The "super tutor" isn't some distant fantasy - millions already use ChatGPT, Claude and similar tools daily for personalized learning. They're imperfect but genuinely helpful for programming, languages, math, and countless other topics.
Look at what happened with YouTube: millions of people transformed themselves into programmers, musicians, mechanics, and countless other professions through free video tutorials. Khan Academy revolutionized math education. Coursera and edX brought university courses to anyone with internet. This wasn't utopian thinking - it was practical technology solving real educational problems at scale.
What's different now is that LLMs enable the missing piece: personalization. The one-on-one adaptive experience that was previously limited to those who could afford human tutors at $50-100/hour is now available to anyone at negligible marginal cost.
Your skepticism about cancer applications too ignores the technological trajectory we've been on for decades. Just as YouTube and online platforms democratized education, technology has been steadily dismantling bottlenecks in medical research.
The human genome project initially cost $3 billion and took 13 years. Today you can sequence a genome for under $1,000 in days. This wasn't utopian thinking; it was technological progress following its natural course.
Think what LLMs will do here.","created_at":"2025-05-18T16:49:58Z","created_at_i":1747586998,"objectID":"44022635","parent_id":44022349,"story_id":44016621,"story_title":"LLMs are more persuasive than incentivized human persuaders","story_url":"https://arxiv.org/abs/2505.09662","updated_at":"2025-05-18T21:15:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jillesvangurp"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"Spot on. We struggle to define what we don't understand. And with AGI, that would include understanding ourselves. And when people start dragging in philosophy, religion, etc. you kind of know it's one of those things that definitely don't have much consensus.
That doesn't mean it's all nonsense. I think objectively we are getting quite a bit of emergent definitions that emerge from established fact and technology that kind of narrow this down. LLMs seem part of the solution on a path to some form of artificial intelligence that can keep up with us and that we might struggle to keep up with. But LLMs are not the whole solution. Though what they can't do keeps shifting. From glorified autocomplete to solving mathematical problems that were previously not solved. In the space of less than 3 years since the launch of chat GPT in November 2022. If the math wasn't solved before, there has to be a bit more to it than a glorified autocomplete.
Of course there are also counter points to that. But it at least challenges the notion that LLMs can't come up with new stuff. Even the notion that they can is deeply upsetting to some people because it challenges their world views. This debate is as much about that as it is about coming up with workable definitions.
LLMs are obviously lacking in a lot of ways. Perhaps the most obvious thing is that after their training is completed, they stop learning and adapting. They have no memory beyond their prompts and what they learned during training. A chat gpt conversation is just a large prompt with the entire history of the conversation and some clever engineering to load that into GPU memory relatively quickly. There are a lot of party tricks that involve elaborate system prompts and other workarounds.
The ability to remember and learn seems pretty fundamental to AGI, whatever that is. It's also a thing that doesn't sound like it's unsolvable. Some kind of indexed DB external to the AI for RAG kind of works but it does not solve the learning problem. And it shifts the problem to how to query and filter the data, which might not be optimal. And it's more similar to us using Google to find something than it is to us remembering information.
Also, it's not like we're particularly good at cramming large amounts of information down or learning. Learning is a slow process. And we kind of get set in our ways as we age. Meaning we're reluctant to learn more stuff and less capable of doing so. That suggests that even modest improvements might have dramatic results. LLMs + some short term memory and ability to learn might end up being a lot more useful.
I like duck typing in programming and that's also my mental model for AGIs. The Turing test is kind of obsolete. But if it quacks like duck and walks like a duck, it probably is a duck. Turing was onto something. Once we have a hard time telling apart the average person in society from an AI over days/weeks/years of interaction, we'll have AGIs. I think that's doable. But I'm not an ML engineer. A lot of those seem to believe it's doable too though.
The rest of society in the form of self appointed arm chair professors, religious leaders, eminent philosophers, and (too put this politely) lesser qualified individuals with strong opinions makes a lot of confused noises. But it does not generate a lot of definitions that have broad consensus."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"AGI Overhyped?"}},"_tags":["comment","author_jillesvangurp","story_45049265"],"author":"jillesvangurp","children":[45054257],"comment_text":"Spot on. We struggle to define what we don't understand. And with AGI, that would include understanding ourselves. And when people start dragging in philosophy, religion, etc. you kind of know it's one of those things that definitely don't have much consensus.
That doesn't mean it's all nonsense. I think objectively we are getting quite a bit of emergent definitions that emerge from established fact and technology that kind of narrow this down. LLMs seem part of the solution on a path to some form of artificial intelligence that can keep up with us and that we might struggle to keep up with. But LLMs are not the whole solution. Though what they can't do keeps shifting. From glorified autocomplete to solving mathematical problems that were previously not solved. In the space of less than 3 years since the launch of chat GPT in November 2022. If the math wasn't solved before, there has to be a bit more to it than a glorified autocomplete.
Of course there are also counter points to that. But it at least challenges the notion that LLMs can't come up with new stuff. Even the notion that they can is deeply upsetting to some people because it challenges their world views. This debate is as much about that as it is about coming up with workable definitions.
LLMs are obviously lacking in a lot of ways. Perhaps the most obvious thing is that after their training is completed, they stop learning and adapting. They have no memory beyond their prompts and what they learned during training. A chat gpt conversation is just a large prompt with the entire history of the conversation and some clever engineering to load that into GPU memory relatively quickly. There are a lot of party tricks that involve elaborate system prompts and other workarounds.
The ability to remember and learn seems pretty fundamental to AGI, whatever that is. It's also a thing that doesn't sound like it's unsolvable. Some kind of indexed DB external to the AI for RAG kind of works but it does not solve the learning problem. And it shifts the problem to how to query and filter the data, which might not be optimal. And it's more similar to us using Google to find something than it is to us remembering information.
Also, it's not like we're particularly good at cramming large amounts of information down or learning. Learning is a slow process. And we kind of get set in our ways as we age. Meaning we're reluctant to learn more stuff and less capable of doing so. That suggests that even modest improvements might have dramatic results. LLMs + some short term memory and ability to learn might end up being a lot more useful.
I like duck typing in programming and that's also my mental model for AGIs. The Turing test is kind of obsolete. But if it quacks like duck and walks like a duck, it probably is a duck. Turing was onto something. Once we have a hard time telling apart the average person in society from an AI over days/weeks/years of interaction, we'll have AGIs. I think that's doable. But I'm not an ML engineer. A lot of those seem to believe it's doable too though.
The rest of society in the form of self appointed arm chair professors, religious leaders, eminent philosophers, and (too put this politely) lesser qualified individuals with strong opinions makes a lot of confused noises. But it does not generate a lot of definitions that have broad consensus.","created_at":"2025-08-28T08:28:30Z","created_at_i":1756369710,"objectID":"45049811","parent_id":45049444,"story_id":45049265,"story_title":"AGI Overhyped?","updated_at":"2026-03-05T22:32:54Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kavaivaleri"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"- AllenNLP Interpret: A library for interpreting and visualizing LLM predictions, assisting in model explanation and debugging. It works for any model of your choice.
- LangKit: An open-source toolkit for monitoring Large Language Models (LLMs). Features include assessing text quality and relevance, hallucinations check, sentiment and toxicity analysis.
- BERTViz: Specifically designed for visualizing and interpreting BERT-based LLMs. Helps visualize attention in NLP Models (BERT, GPT2, BART, etc.).
- SHAP (SHapley Additive exPlanations): A game theoretic approach to explain the output of any machine learning model. Allows users to use models from the transformers library by HuggingFace.
- AI Fairness 360: An extensible open-source toolkit can help you examine, report, and mitigate discrimination and bias in machine learning models throughout the AI application lifecycle
- Prometheus: An open-source monitoring toolkit for collecting and querying metrics from LLMs in real time.
- Grafana: Integrates with tools like Prometheus and Elasticsearch to provide visualization and analysis of LLM metrics and logs.
Learn more about LLM monitoring and observability at https://www.turingpost.com/p/monitoring"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["llm"],"value":"Open-source tools for LLM monitoring and observability"}},"_tags":["story","author_kavaivaleri","story_39215123","ask_hn"],"author":"kavaivaleri","children":[39215276,39222543,39226167,39259642,39308091],"created_at":"2024-02-01T12:10:35Z","created_at_i":1706789435,"num_comments":9,"objectID":"39215123","points":4,"story_id":39215123,"story_text":"- AllenNLP Interpret: A library for interpreting and visualizing LLM predictions, assisting in model explanation and debugging. It works for any model of your choice.
- LangKit: An open-source toolkit for monitoring Large Language Models (LLMs). Features include assessing text quality and relevance, hallucinations check, sentiment and toxicity analysis.
- BERTViz: Specifically designed for visualizing and interpreting BERT-based LLMs. Helps visualize attention in NLP Models (BERT, GPT2, BART, etc.).
- SHAP (SHapley Additive exPlanations): A game theoretic approach to explain the output of any machine learning model. Allows users to use models from the transformers library by HuggingFace.
- AI Fairness 360: An extensible open-source toolkit can help you examine, report, and mitigate discrimination and bias in machine learning models throughout the AI application lifecycle
- Prometheus: An open-source monitoring toolkit for collecting and querying metrics from LLMs in real time.
- Grafana: Integrates with tools like Prometheus and Elasticsearch to provide visualization and analysis of LLM metrics and logs.
Learn more about LLM monitoring and observability at https://www.turingpost.com/p/monitoring","title":"Open-source tools for LLM monitoring and observability","updated_at":"2024-09-20T16:13:36Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hodgehog11"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"As someone who has students that work in deep learning, I can say that it is unwise to approach deep learning in the same way as traditional ML. Most classical methods are strongly mathematically motivated and have excellent theory to accompany them. Deep learning is still alchemy; it is a matter of experience, trying things out and getting a feel for how the pieces fit together in a modular format. Once you are experienced with the common building blocks, you can develop an intuition for how they might be improved.
I would start with training a basic MLP on tabular data. Then switch to CNNs: LeNet, VGG, then ResNet. Understand each of the new blocks that are incorporated into each architecture and how they improve stability and training efficiency. There are good PyTorch tutorials for these. Use these as a playground to understand what each of the training knobs do. Look at how their implicit biases induce double descent; this should give you confidence that overfitting is rarely an issue anymore. Give finetuning a try by taking a pretrained ResNet on ImageNet, adding layers to the start and end, and training only these to adapt the model to another image dataset. This should demonstrate the power of finetuning and why pretrained models are so powerful.
Next, briefly consider a tutorial on LSTMs, recognizing the exploding and vanishing gradient problems and the traditional challenges with sequential data.
Then move to transformers. Work with language first, starting from Andrej Karpathy's excellent YouTube tutorials. Train the model in full for a bit, then see about using an existing GPT2 checkpoint. Try adapting NanoGPT to a mathematical dataset as an exercise. Then take a look at llm.c to see how to really improve performance.
Finally, take a look at ViT and DETR. Use pretrained models and finetune them on smaller datasets again.
By this point, you should have a good grounding to start reading much of the surrounding literature and understand them. You should also understand that models are never built from scratch anymore, and every model is a collection of individual pieces built elsewhere for a particular purpose."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Gemma 3 270M re-implemented in pure PyTorch for local tinkering"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/rasbt/LLMs-from-scratch/tree/main/ch05/12_gemma3"}},"_tags":["comment","author_hodgehog11","story_44962059"],"author":"hodgehog11","children":[44981447],"comment_text":"As someone who has students that work in deep learning, I can say that it is unwise to approach deep learning in the same way as traditional ML. Most classical methods are strongly mathematically motivated and have excellent theory to accompany them. Deep learning is still alchemy; it is a matter of experience, trying things out and getting a feel for how the pieces fit together in a modular format. Once you are experienced with the common building blocks, you can develop an intuition for how they might be improved.
I would start with training a basic MLP on tabular data. Then switch to CNNs: LeNet, VGG, then ResNet. Understand each of the new blocks that are incorporated into each architecture and how they improve stability and training efficiency. There are good PyTorch tutorials for these. Use these as a playground to understand what each of the training knobs do. Look at how their implicit biases induce double descent; this should give you confidence that overfitting is rarely an issue anymore. Give finetuning a try by taking a pretrained ResNet on ImageNet, adding layers to the start and end, and training only these to adapt the model to another image dataset. This should demonstrate the power of finetuning and why pretrained models are so powerful.
Next, briefly consider a tutorial on LSTMs, recognizing the exploding and vanishing gradient problems and the traditional challenges with sequential data.
Then move to transformers. Work with language first, starting from Andrej Karpathy's excellent YouTube tutorials. Train the model in full for a bit, then see about using an existing GPT2 checkpoint. Try adapting NanoGPT to a mathematical dataset as an exercise. Then take a look at llm.c to see how to really improve performance.
Finally, take a look at ViT and DETR. Use pretrained models and finetune them on smaller datasets again.
By this point, you should have a good grounding to start reading much of the surrounding literature and understand them. You should also understand that models are never built from scratch anymore, and every model is a collection of individual pieces built elsewhere for a particular purpose.","created_at":"2025-08-21T06:32:55Z","created_at_i":1755757975,"objectID":"44969705","parent_id":44968323,"story_id":44962059,"story_title":"Gemma 3 270M re-implemented in pure PyTorch for local tinkering","story_url":"https://github.com/rasbt/LLMs-from-scratch/tree/main/ch05/12_gemma3","updated_at":"2026-03-05T22:32:18Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"TeMPOraL"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["adaptive","learning","llm","tutoring"],"value":"The best voice assistant I ever had was the one I DIY-ed around 2007, using Microsoft Speech API and a cheap piezoelectric mike I soldered to a long cable, hung off the wardrobe, and plugged into PC. The code itself was a mashup of MS SAPI demo of "controlled language" interface and some tutorials for how to control WinAMP with WM_USER messages in WinAPI. I designed a little tree of commands, maybe 3 level deep, wrote the magic XML for it, some trivial C++ logic for driving the voice recognizer and reacting to identified commands (including one that held the recognizer two levels deep in the command tree, so I could issue multiple commands from a subtree without having to repeat two extra words for each).
The result of this couple afternoons of working on this (instead of learning for my maturity exams, as I was supposed to), was a system that, I kid you not, was more reliable and delivered more value to me than any of the current voice assistants. For one, its recognition was flawless. The typical interaction would look like:
$ Computer!\n > <appropriate beep from Star Trek: TNG, because of course\n doing this was 90% of my reason for building the program>\n $ Music, Playlist Alpha\n > <appropriate confirmation beep, WinAMP begins to play>\nHere's the thing: it worked flawlessly. It heard me across the room. It heard me through music so loud that it was uncomfortable to talk in. It never self-triggered (except that one person who managed to make a swear word be read as the wake word, a single case out of many who tried). It worked fast - I could complete the whole command chain in less time than Google Assistant takes to start listening after "OK Google". The secret? Constrained grammar and training.
In order to use speech recognition in Windows back then, you had to turn it on and let it analyze a sample of your voice (offline! those were the days!), based on a recording of you reading some calibration text it gave you. This process was additive - you could repeat it to improve recognition accuracy. But a little known fact was that you could also supply your own text - and that was the other half that made the magic happen.
I created myself a training text, consisting of individual command words and their sequences, and trained the Windows speech recognition on it multiple times, under varying conditions. Specifically, I run:
{three locations in the room} x ({no background} + ({classical music, pop music, whatever was on FM radio} x {quiet playback, normal playback, very loud playback}))
training sessions. That's 30 sessions of repeating the same text. Each one took maybe a minute or less, so I was done with it in about an hour. And after that training, no matter where I was standing in the room and what I was doing, the voice control system worked with near-zero false positives and near-zero false negatives. I say "near" because I had maybe two or three cases of each, over months of continued use. And yes, I could play music so loud you couldn't talk in the room, and I could scream out commands, outshouting the music, and it would work. Try that with Google Assistant.
To recap: I had a system I hacked together in couple evenings, whose software was a relatively small tweak to a default example project (but done with love!) and hardware was hand-soldered from cheapest, locally-sourced parts, that did everything I wanted from a voice assistant, did it flawlessly, much faster than any of the voice assistants on the market today, completely off-line, in 2007, on a mid-range PC, without noticeably taxing its resources. This is why I occasionally rant that voice assistants are bloated and done backwards - all because they're designed to suit vendor needs first, user needs second.
--------
But hey, I know a way Google, Apple, Samsung (!) et al. could fix the shitty performance of their voice assistants and dictation software. They need to fine-tune a LLM on a dataset made of target words/sentences, and transcripts of them being misheard in great many ways. Then they need to feed the output of their voice-to-text pipeline through that LLM, so it can correct the text wholesale. That, or maybe, you know, do whatever Microsoft was doing in 2007 that made dictation work well and offline."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Natural language is an unnatural interface"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://varunshenoy.substack.com/p/natural-language-is-an-unnatural"}},"_tags":["comment","author_TeMPOraL","story_36500945"],"author":"TeMPOraL","children":[36504162],"comment_text":"The best voice assistant I ever had was the one I DIY-ed around 2007, using Microsoft Speech API and a cheap piezoelectric mike I soldered to a long cable, hung off the wardrobe, and plugged into PC. The code itself was a mashup of MS SAPI demo of "controlled language" interface and some tutorials for how to control WinAMP with WM_USER messages in WinAPI. I designed a little tree of commands, maybe 3 level deep, wrote the magic XML for it, some trivial C++ logic for driving the voice recognizer and reacting to identified commands (including one that held the recognizer two levels deep in the command tree, so I could issue multiple commands from a subtree without having to repeat two extra words for each).
The result of this couple afternoons of working on this (instead of learning for my maturity exams, as I was supposed to), was a system that, I kid you not, was more reliable and delivered more value to me than any of the current voice assistants. For one, its recognition was flawless. The typical interaction would look like:
$ Computer!\n > <appropriate beep from Star Trek: TNG, because of course\n doing this was 90% of my reason for building the program>\n $ Music, Playlist Alpha\n > <appropriate confirmation beep, WinAMP begins to play>\nHere's the thing: it worked flawlessly. It heard me across the room. It heard me through music so loud that it was uncomfortable to talk in. It never self-triggered (except that one person who managed to make a swear word be read as the wake word, a single case out of many who tried). It worked fast - I could complete the whole command chain in less time than Google Assistant takes to start listening after "OK Google". The secret? Constrained grammar and training.
In order to use speech recognition in Windows back then, you had to turn it on and let it analyze a sample of your voice (offline! those were the days!), based on a recording of you reading some calibration text it gave you. This process was additive - you could repeat it to improve recognition accuracy. But a little known fact was that you could also supply your own text - and that was the other half that made the magic happen.
I created myself a training text, consisting of individual command words and their sequences, and trained the Windows speech recognition on it multiple times, under varying conditions. Specifically, I run:
{three locations in the room} x ({no background} + ({classical music, pop music, whatever was on FM radio} x {quiet playback, normal playback, very loud playback}))
training sessions. That's 30 sessions of repeating the same text. Each one took maybe a minute or less, so I was done with it in about an hour. And after that training, no matter where I was standing in the room and what I was doing, the voice control system worked with near-zero false positives and near-zero false negatives. I say "near" because I had maybe two or three cases of each, over months of continued use. And yes, I could play music so loud you couldn't talk in the room, and I could scream out commands, outshouting the music, and it would work. Try that with Google Assistant.
To recap: I had a system I hacked together in couple evenings, whose software was a relatively small tweak to a default example project (but done with love!) and hardware was hand-soldered from cheapest, locally-sourced parts, that did everything I wanted from a voice assistant, did it flawlessly, much faster than any of the voice assistants on the market today, completely off-line, in 2007, on a mid-range PC, without noticeably taxing its resources. This is why I occasionally rant that voice assistants are bloated and done backwards - all because they're designed to suit vendor needs first, user needs second.
--------
But hey, I know a way Google, Apple, Samsung (!) et al. could fix the shitty performance of their voice assistants and dictation software. They need to fine-tune a LLM on a dataset made of target words/sentences, and transcripts of them being misheard in great many ways. Then they need to feed the output of their voice-to-text pipeline through that LLM, so it can correct the text wholesale. That, or maybe, you know, do whatever Microsoft was doing in 2007 that made dictation work well and offline.","created_at":"2023-06-28T08:34:13Z","created_at_i":1687941253,"objectID":"36503859","parent_id":36503488,"story_id":36500945,"story_title":"Natural language is an unnatural interface","story_url":"https://varunshenoy.substack.com/p/natural-language-is-an-unnatural","updated_at":"2024-09-20T14:21:36Z"}],"hitsPerPage":20,"nbHits":9,"nbPages":1,"page":0,"params":"query=adaptive+learning+LLM+tutoring&advancedSyntax=true&analyticsTags=backend","processingTimeMS":33,"processingTimingsMS":{"_request":{"queue":2,"roundTrip":23},"afterFetch":{"format":{"highlighting":3,"total":3},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":2},"total":2},"fetch":{"query":24,"scanning":5,"total":30},"total":33},"query":"adaptive learning LLM tutoring","serverTimeMS":39}