If you use a thread per connection (or green threads like Go), you don't also need async. If you have async (eg nodejs), you can get great performance without threads. You're right that they can also be combined - either within a single process (like tokio in rust). Or via multi-process configurations (eg one nodejs instance per core, all behind nginx). But they don't need to be. Go (green threads) and Nodejs (async, single threads) both work well.
Of course we're comparing them. We all want to know who wore it better"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"What async promised and what it delivered"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://causality.blog/essays/what-async-promised/"}},"_tags":["comment","author_josephg","story_47859442"],"author":"josephg","children":[47908446],"comment_text":"Of course I'm comparing them. Threading and async are two solutions to the same problem: How do you write high performance event driven systems like network services? How do you solve the C10K problem (or more recently the C10M problem)?
If you use a thread per connection (or green threads like Go), you don't also need async. If you have async (eg nodejs), you can get great performance without threads. You're right that they can also be combined - either within a single process (like tokio in rust). Or via multi-process configurations (eg one nodejs instance per core, all behind nginx). But they don't need to be. Go (green threads) and Nodejs (async, single threads) both work well.
Of course we're comparing them. We all want to know who wore it better","created_at":"2026-04-26T07:09:01Z","created_at_i":1777187341,"objectID":"47908087","parent_id":47905632,"story_id":47859442,"story_title":"What async promised and what it delivered","story_url":"https://causality.blog/essays/what-async-promised/","updated_at":"2026-04-26T08:21:36Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"luew"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"So we're a small inference provider, launched publicly two weeks ago and have seen a crazy demand of growth.
I reached out to a lot of other inference providers such as fireworks, togetherAI, simpliAI etc and started asking them their growth and what they are seeing in this space / what they predict we will see over this year.
I was told by a higher up at fireworks that on average since January the space as a whole has had 10% week over week growth -- this type of explosive growth feels unreal. I honestly expect a price squeeze later this year, no one can keep up.
I think we're starting to see this with GPU prices -- h100s have gone from 1.30 on demand to 1.90 about, h200 nodes are all sold out with 3+ month waitlists. B300s have near a year waitlist from the quotes I got. Insane market."},"title":{"matchLevel":"none","matchedWords":[],"value":"Open-Source Inference is growing 10% week over week this year"}},"_tags":["story","author_luew","story_47906463","ask_hn"],"author":"luew","children":[47906976],"created_at":"2026-04-26T01:33:31Z","created_at_i":1777167211,"num_comments":1,"objectID":"47906463","points":2,"story_id":47906463,"story_text":"So we're a small inference provider, launched publicly two weeks ago and have seen a crazy demand of growth.
I reached out to a lot of other inference providers such as fireworks, togetherAI, simpliAI etc and started asking them their growth and what they are seeing in this space / what they predict we will see over this year.
I was told by a higher up at fireworks that on average since January the space as a whole has had 10% week over week growth -- this type of explosive growth feels unreal. I honestly expect a price squeeze later this year, no one can keep up.
I think we're starting to see this with GPU prices -- h100s have gone from 1.30 on demand to 1.90 about, h200 nodes are all sold out with 3+ month waitlists. B300s have near a year waitlist from the quotes I got. Insane market.","title":"Open-Source Inference is growing 10% week over week this year","updated_at":"2026-04-26T03:17:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Zta77"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Proxmox is great. I use it for experimenting with Lightwhale swarm clusters locally; I'm currently running 8 Lightwhale worker nodes with 512MB RAM each and no disk, and it's super fast and fits in my pocket.
I don't consider Lightwhale as an alternative to Proxmox. In fact, how do you even run a Docker container in Proxmox? Without booting Lightwhale in a VM first, I mean? ;)"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: I've built a nice home server OS"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://lightwhale.asklandd.dk/"}},"_tags":["comment","author_Zta77","story_47896163"],"author":"Zta77","comment_text":"Proxmox is great. I use it for experimenting with Lightwhale swarm clusters locally; I'm currently running 8 Lightwhale worker nodes with 512MB RAM each and no disk, and it's super fast and fits in my pocket.
I don't consider Lightwhale as an alternative to Proxmox. In fact, how do you even run a Docker container in Proxmox? Without booting Lightwhale in a VM first, I mean? ;)","created_at":"2026-04-25T21:49:16Z","created_at_i":1777153756,"objectID":"47904855","parent_id":47897488,"story_id":47896163,"story_title":"Show HN: I've built a nice home server OS","story_url":"https://lightwhale.asklandd.dk/","updated_at":"2026-04-25T21:52:36Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bmitc"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"I was not very descriptive, but I was referring to the next layer up of building blocks. Instead of text, we could also express things in hybrid ways with text but also visual nodes that can carry more dense information. The usual response is that those things don't work with text-based tools, but that's my point. Text based tools needed invention and decades of refinement, and they're still not all that great."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Plain text has been around for decades and it\u2019s here to stay"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://unsung.aresluna.org/plain-text-has-been-around-for-decades-and-its-here-to-stay/"}},"_tags":["comment","author_bmitc","story_47897681"],"author":"bmitc","comment_text":"I was not very descriptive, but I was referring to the next layer up of building blocks. Instead of text, we could also express things in hybrid ways with text but also visual nodes that can carry more dense information. The usual response is that those things don't work with text-based tools, but that's my point. Text based tools needed invention and decades of refinement, and they're still not all that great.","created_at":"2026-04-25T17:59:51Z","created_at_i":1777139991,"objectID":"47903259","parent_id":47902569,"story_id":47897681,"story_title":"Plain text has been around for decades and it\u2019s here to stay","story_url":"https://unsung.aresluna.org/plain-text-has-been-around-for-decades-and-its-here-to-stay/","updated_at":"2026-04-25T18:11:03Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"CTSuwan"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"A few years ago, I was a delivery driver in Bangkok. I saw firsthand how inefficient algorithms stressed out drivers. At that time, I didn't even know what "NP-hard" meant\u2014I just knew the system could be better. So, I started building.
The Journey of an Outsider:\nI have no CS background. I hold a vocational diploma in Goldsmithing from 20 years ago. Before this, I was unemployed and had no PC. My only tool was a $100 Android smartphone (3,000 THB).
I spent 16 hours a day architecting the logic via Pydroid 3. Because I didn't know standard optimization libraries existed, I designed my own deterministic logic architecture from the ground up. I just thought that was how software was built.
The Technical Skepticism:\nWhen I shared my work locally, the skepticism was purely technical. People couldn't believe a standard Snapdragon environment could solve 10,000-node VRP instances without runtime explosions, doubting mobile hardware could handle an NP-hard problem of this scale.
The Result:\nBy relying purely on deterministic, axiomatic logic rather than standard metaheuristics, the engine (GSL Solver) now handles up to 10,000 nodes with stable execution across standard benchmarks (CVRP, VRPTW, MDVRP).
I\u2019ve kept the benchmark outputs transparent for inspection: https://github.com/CT1-deMo-goG/CT1-deMo-goG
You can run the live deterministic engine here: https://gsl-solver.com
P.S. Even the front-end website was built entirely on that same smartphone using Acode. I'd love to hear your thoughts on the architectural approach of building solvers entirely from scratch without standard libraries."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN:I built a deterministic 10k-node VRP solver on a $100 phone"}},"_tags":["story","author_CTSuwan","story_47902994","show_hn"],"author":"CTSuwan","children":[47903007],"created_at":"2026-04-25T17:21:26Z","created_at_i":1777137686,"num_comments":0,"objectID":"47902994","points":2,"story_id":47902994,"story_text":"A few years ago, I was a delivery driver in Bangkok. I saw firsthand how inefficient algorithms stressed out drivers. At that time, I didn't even know what "NP-hard" meant\u2014I just knew the system could be better. So, I started building.
The Journey of an Outsider:\nI have no CS background. I hold a vocational diploma in Goldsmithing from 20 years ago. Before this, I was unemployed and had no PC. My only tool was a $100 Android smartphone (3,000 THB).
I spent 16 hours a day architecting the logic via Pydroid 3. Because I didn't know standard optimization libraries existed, I designed my own deterministic logic architecture from the ground up. I just thought that was how software was built.
The Technical Skepticism:\nWhen I shared my work locally, the skepticism was purely technical. People couldn't believe a standard Snapdragon environment could solve 10,000-node VRP instances without runtime explosions, doubting mobile hardware could handle an NP-hard problem of this scale.
The Result:\nBy relying purely on deterministic, axiomatic logic rather than standard metaheuristics, the engine (GSL Solver) now handles up to 10,000 nodes with stable execution across standard benchmarks (CVRP, VRPTW, MDVRP).
I\u2019ve kept the benchmark outputs transparent for inspection: https://github.com/CT1-deMo-goG/CT1-deMo-goG
You can run the live deterministic engine here: https://gsl-solver.com
P.S. Even the front-end website was built entirely on that same smartphone using Acode. I'd love to hear your thoughts on the architectural approach of building solvers entirely from scratch without standard libraries.","title":"Show HN:I built a deterministic 10k-node VRP solver on a $100 phone","updated_at":"2026-04-25T18:42:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hnlmorg"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Isn\u2019t that only relevant for network topologies that rely heavily on broadcasting to multiple nodes. Eg token ring, WiFi and powerline adapters?
For regular Ethernet, the switch will have a table of which IPs are on which NIC and thus can dynamically send packets at the right transmission protocols supported by those NICs without degrading the service of other NICs."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"New 10 GbE USB adapters are cooler, smaller, cheaper"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.jeffgeerling.com/blog/2026/new-10-gbe-usb-adapters-cooler-smaller-cheaper/"}},"_tags":["comment","author_hnlmorg","story_47899053"],"author":"hnlmorg","children":[47900377],"comment_text":"Isn\u2019t that only relevant for network topologies that rely heavily on broadcasting to multiple nodes. Eg token ring, WiFi and powerline adapters?
For regular Ethernet, the switch will have a table of which IPs are on which NIC and thus can dynamically send packets at the right transmission protocols supported by those NICs without degrading the service of other NICs.","created_at":"2026-04-25T10:32:41Z","created_at_i":1777113161,"objectID":"47900320","parent_id":47900114,"story_id":47899053,"story_title":"New 10 GbE USB adapters are cooler, smaller, cheaper","story_url":"https://www.jeffgeerling.com/blog/2026/new-10-gbe-usb-adapters-cooler-smaller-cheaper/","updated_at":"2026-04-25T19:40:20Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"stealthtsdb"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"I built a browser-only studio for designing and orchestrating MCP agent systems for development and experimental purposes. The whole stack \u2014 tool authoring, multi-agent orchestration, RAG, code execution \u2014 runs from a single static HTML file via WebAssembly. No backend.
The bet: WASM is a hard sandbox for free. When you generate tools with an LLM (or write them by hand), the studio AST-validates the source, registers it lazily, and JIT-compiles into Pyodide on first call. SQL tools run in DuckDB-WASM in a Web Worker. The built-in RAG uses Xenova/all-MiniLM-L6-v2 via Transformers.js for on-device embeddings. Nothing leaves the browser; close the tab and the stack is gone. The WASM boundary is what makes it safe to execute LLM-generated code locally \u2014 no Docker, no per-tenant container, no server.
Above the tool layer sits an agentic system with 10 orchestration strategies:
- Supervisor (router \u2192 1 expert)\n- Mixture of Experts (parallel + synthesizer)\n- Sequential Pipeline\n- Plan & Execute (planner decomposes, workers execute)\n- Swarm (peer handoffs)\n- Debate (contestants + judge)\n- Reflection (actor + critic loop)\n- Hierarchical (manager delegates via ask_<persona> tools)\n- Round-Robin (panel + moderator)\n- Map-Reduce (splitter \u2192 parallel \u2192 aggregator)
You build a team visually: drag tool chips onto persona nodes on a service graph, pick a strategy, and the topology reshapes to match. Each persona auto-registers as an MCP tool (ask_<name>), plus an agent_chat(query, strategy?) meta tool. A bundled Node bridge speaks stdio to Claude Desktop and WebSocket to your tab \u2014 your browser becomes an MCP server.
When you're done, Export gives you a real Python MCP server: server.py, agentic.py, tools/*.py, Dockerfile, requirements.txt, .env.example. The exported agentic.py is a faithful Python port of the same orchestration logic running in the browser, so the deployable artifact behaves identically to the prototype.
Also shipped: Project Packs. Export the whole project as a single .agentpack.json. Auto-detects required external services (OpenAI, GitHub, Stripe, Anthropic, Slack, Notion, Linear, etc.) by scanning tool source for os.environ.get(...) and cross-referencing against the network allowlist. Recipients get an import wizard that prompts for credentials. Manifests are reviewable, sharable, and never carry secrets.
Some things I'm honestly uncertain about:
- 10 strategies might be too many. My guess is most users only need Supervisor, Mixture of Experts, and Debate. Open to data on which ones actually pull weight.\n- Browser cold-starts (Pyodide warm-up on first load) are a real UX hit despite aggressive caching.\n- bridge.js is the only non-browser piece. A hosted variant is the obvious next step.
Built with Pyodide, DuckDB-WASM, Transformers.js, OpenAI Chat Completions (or a local Qwen 1.5 0.5B running in-browser via Transformers for fully offline mode). ~5K lines of HTML/CSS/JS in one file.
Genuinely curious whether running this much LLM-generated code in a browser tab feels reasonable to you, or quietly terrifying."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Agent MCP Studio \u2013 build multi-agent MCP systems in a browser tab"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.agentmcp.studio"}},"_tags":["story","author_stealthtsdb","story_47899375","show_hn"],"author":"stealthtsdb","children":[47901353,47902781,47900007,47899406],"created_at":"2026-04-25T07:06:31Z","created_at_i":1777100791,"num_comments":3,"objectID":"47899375","points":11,"story_id":47899375,"story_text":"I built a browser-only studio for designing and orchestrating MCP agent systems for development and experimental purposes. The whole stack \u2014 tool authoring, multi-agent orchestration, RAG, code execution \u2014 runs from a single static HTML file via WebAssembly. No backend.
The bet: WASM is a hard sandbox for free. When you generate tools with an LLM (or write them by hand), the studio AST-validates the source, registers it lazily, and JIT-compiles into Pyodide on first call. SQL tools run in DuckDB-WASM in a Web Worker. The built-in RAG uses Xenova/all-MiniLM-L6-v2 via Transformers.js for on-device embeddings. Nothing leaves the browser; close the tab and the stack is gone. The WASM boundary is what makes it safe to execute LLM-generated code locally \u2014 no Docker, no per-tenant container, no server.
Above the tool layer sits an agentic system with 10 orchestration strategies:
- Supervisor (router \u2192 1 expert)\n- Mixture of Experts (parallel + synthesizer)\n- Sequential Pipeline\n- Plan & Execute (planner decomposes, workers execute)\n- Swarm (peer handoffs)\n- Debate (contestants + judge)\n- Reflection (actor + critic loop)\n- Hierarchical (manager delegates via ask_<persona> tools)\n- Round-Robin (panel + moderator)\n- Map-Reduce (splitter \u2192 parallel \u2192 aggregator)
You build a team visually: drag tool chips onto persona nodes on a service graph, pick a strategy, and the topology reshapes to match. Each persona auto-registers as an MCP tool (ask_<name>), plus an agent_chat(query, strategy?) meta tool. A bundled Node bridge speaks stdio to Claude Desktop and WebSocket to your tab \u2014 your browser becomes an MCP server.
When you're done, Export gives you a real Python MCP server: server.py, agentic.py, tools/*.py, Dockerfile, requirements.txt, .env.example. The exported agentic.py is a faithful Python port of the same orchestration logic running in the browser, so the deployable artifact behaves identically to the prototype.
Also shipped: Project Packs. Export the whole project as a single .agentpack.json. Auto-detects required external services (OpenAI, GitHub, Stripe, Anthropic, Slack, Notion, Linear, etc.) by scanning tool source for os.environ.get(...) and cross-referencing against the network allowlist. Recipients get an import wizard that prompts for credentials. Manifests are reviewable, sharable, and never carry secrets.
Some things I'm honestly uncertain about:
- 10 strategies might be too many. My guess is most users only need Supervisor, Mixture of Experts, and Debate. Open to data on which ones actually pull weight.\n- Browser cold-starts (Pyodide warm-up on first load) are a real UX hit despite aggressive caching.\n- bridge.js is the only non-browser piece. A hosted variant is the obvious next step.
Built with Pyodide, DuckDB-WASM, Transformers.js, OpenAI Chat Completions (or a local Qwen 1.5 0.5B running in-browser via Transformers for fully offline mode). ~5K lines of HTML/CSS/JS in one file.
Genuinely curious whether running this much LLM-generated code in a browser tab feels reasonable to you, or quietly terrifying.","title":"Show HN: Agent MCP Studio \u2013 build multi-agent MCP systems in a browser tab","updated_at":"2026-04-25T23:07:20Z","url":"https://www.agentmcp.studio"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"knolan"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"I love that Blender can be a powerful scientific tool. Unlike a lot of software build for research, Blender is fast and user friendly, even if it has a learning curve. The addition of geometry nodes adds so much more capability too.
I\u2019ve used it to help students visualise topologies for meshing in command line CFD tools. It\u2019s also great for motion tracking objects in video footage. I\u2019ve even used it to simulate various camera systems.
Unfortunately the videos aren\u2019t working on my iPhone. I\u2019ll try on desktop later."},"story_title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Cosmology with Geometry Nodes"},"story_url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"https://www.blender.org/user-stories/cosmology-with-geometry-nodes/"}},"_tags":["comment","author_knolan","story_47897706"],"author":"knolan","children":[47901719],"comment_text":"I love that Blender can be a powerful scientific tool. Unlike a lot of software build for research, Blender is fast and user friendly, even if it has a learning curve. The addition of geometry nodes adds so much more capability too.
I\u2019ve used it to help students visualise topologies for meshing in command line CFD tools. It\u2019s also great for motion tracking objects in video footage. I\u2019ve even used it to simulate various camera systems.
Unfortunately the videos aren\u2019t working on my iPhone. I\u2019ll try on desktop later.","created_at":"2026-04-25T06:56:59Z","created_at_i":1777100219,"objectID":"47899338","parent_id":47897706,"story_id":47897706,"story_title":"Cosmology with Geometry Nodes","story_url":"https://www.blender.org/user-stories/cosmology-with-geometry-nodes/","updated_at":"2026-04-25T22:15:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"shankysingh"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Cosmology with Geometry Nodes"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"https://www.blender.org/user-stories/cosmology-with-geometry-nodes/"}},"_tags":["story","author_shankysingh","story_47897706"],"author":"shankysingh","children":[47899338,47901888],"created_at":"2026-04-25T01:08:14Z","created_at_i":1777079294,"num_comments":6,"objectID":"47897706","points":106,"story_id":47897706,"title":"Cosmology with Geometry Nodes","updated_at":"2026-04-26T10:27:51Z","url":"https://www.blender.org/user-stories/cosmology-with-geometry-nodes/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ux266478"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"They should probably look at moving away from general purpose hardware for their actual products, and reserve GP hardware for RnD. You don't need frontier nodes to run circles around GPGPUs, an ASIC made with 28nm is more than enough to embarrass an H100 (and much cheaper)
AI is in such desperate need to adopt software-hardware co-development practices, it's infuriating watching the industry drag its feet about it. We are wasting so much electricity and absolutely wrecking the "free" market just because these companies are incentivized to work at an unsustainable breakneck speed in getting shit to market."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Google plans to invest up to $40B in Anthropic"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.bloomberg.com/news/articles/2026-04-24/google-plans-to-invest-up-to-40-billion-in-anthropic"}},"_tags":["comment","author_ux266478","story_47892074"],"author":"ux266478","comment_text":"They should probably look at moving away from general purpose hardware for their actual products, and reserve GP hardware for RnD. You don't need frontier nodes to run circles around GPGPUs, an ASIC made with 28nm is more than enough to embarrass an H100 (and much cheaper)
AI is in such desperate need to adopt software-hardware co-development practices, it's infuriating watching the industry drag its feet about it. We are wasting so much electricity and absolutely wrecking the "free" market just because these companies are incentivized to work at an unsustainable breakneck speed in getting shit to market.","created_at":"2026-04-25T00:35:04Z","created_at_i":1777077304,"objectID":"47897479","parent_id":47895553,"story_id":47892074,"story_title":"Google plans to invest up to $40B in Anthropic","story_url":"https://www.bloomberg.com/news/articles/2026-04-24/google-plans-to-invest-up-to-40-billion-in-anthropic","updated_at":"2026-04-25T14:41:33Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"srean"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"> NNs were largely dismissed
I agree with your larger point but dismissed is rather too strong. They were considered fiddly to train, prone to local minima, long training time, no clear guidelines about what the number of hidden layers and number of nodes ought to be. But for homework (toy) exercises they were still ok.
In comparison, kernel methods gave a better experience over all for large but not super large data sets. Most models had easily obtainable global minimum. Fewer moving parts and very good performance.
It turns out, however, that if you have several orders of magnitude more data, the usual kernels are too simple -- (i) they cannot take advantage of more data after a point and start twiddling the 10th place of decimal of some parameters and (ii) are expensive to train for very large data sets. So bit of a double whammy. Well, there was a third, no hardware acceleration that can compare with GPUs.
Kernels may make a comeback though, you never know. We need to find a way to compose kernels in a user friendly way to increase their modeling capacity. We had a few ways of doing just that but they weren't great. We need a breakthrough to scale them to GPT sized data sets.
In a way DNNs are "design your own kernels using data" whereas kernels came in any color you liked provided it was black (yes there were many types, but it was still a fairly limited catalogue. The killer was that there was no good way of composing them to increase modeling capacity that yielded efficiently trainable kernel machines)"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"There Will Be a Scientific Theory of Deep Learning"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2604.21691"}},"_tags":["comment","author_srean","story_47893779"],"author":"srean","comment_text":"> NNs were largely dismissed
I agree with your larger point but dismissed is rather too strong. They were considered fiddly to train, prone to local minima, long training time, no clear guidelines about what the number of hidden layers and number of nodes ought to be. But for homework (toy) exercises they were still ok.
In comparison, kernel methods gave a better experience over all for large but not super large data sets. Most models had easily obtainable global minimum. Fewer moving parts and very good performance.
It turns out, however, that if you have several orders of magnitude more data, the usual kernels are too simple -- (i) they cannot take advantage of more data after a point and start twiddling the 10th place of decimal of some parameters and (ii) are expensive to train for very large data sets. So bit of a double whammy. Well, there was a third, no hardware acceleration that can compare with GPUs.
Kernels may make a comeback though, you never know. We need to find a way to compose kernels in a user friendly way to increase their modeling capacity. We had a few ways of doing just that but they weren't great. We need a breakthrough to scale them to GPT sized data sets.
In a way DNNs are "design your own kernels using data" whereas kernels came in any color you liked provided it was black (yes there were many types, but it was still a fairly limited catalogue. The killer was that there was no good way of composing them to increase modeling capacity that yielded efficiently trainable kernel machines)","created_at":"2026-04-24T22:45:13Z","created_at_i":1777070713,"objectID":"47896738","parent_id":47896430,"story_id":47893779,"story_title":"There Will Be a Scientific Theory of Deep Learning","story_url":"https://arxiv.org/abs/2604.21691","updated_at":"2026-04-25T15:12:18Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"405nm"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"that map on rmap.world is only showing nodes that run dicoverable=yes in their configurations or something like that.
based upon the announce stream coming through my local node, i am seeing around 14k unique identities advertising over 21k unique application endpoints (destinations) over the course of the past month or so that i\u2019ve been tracking it."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"MeshCore development team splits over trademark dispute and AI-generated code"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://blog.meshcore.io/2026/04/23/the-split"}},"_tags":["comment","author_405nm","story_47878117"],"author":"405nm","comment_text":"that map on rmap.world is only showing nodes that run dicoverable=yes in their configurations or something like that.
based upon the announce stream coming through my local node, i am seeing around 14k unique identities advertising over 21k unique application endpoints (destinations) over the course of the past month or so that i\u2019ve been tracking it.","created_at":"2026-04-24T19:21:49Z","created_at_i":1777058509,"objectID":"47894652","parent_id":47885160,"story_id":47878117,"story_title":"MeshCore development team splits over trademark dispute and AI-generated code","story_url":"https://blog.meshcore.io/2026/04/23/the-split","updated_at":"2026-04-25T10:34:03Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rtmx"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"TBH, I'd say we were there long before LLMs came to 'help'. Software world was in a dreadful state for a decade or so, maybe longer, I'd say \u2014 devs get powerful machines not like normal users, everyone is 'just doing their job', software tested in isolated environments so no one cares about installing a couple of their own 'background services in NodeJS' on a user machine \u2014 not a big deal, yeah? And so on...
At the same time, I see the future being brighter with the help of these coding LLMs \u2014 I personally was not building software for years, focusing on management-like work. Serious coding during 'free time' was just too heavy to lift \u2014 you need time to sleep, eat and do some IRL things too...
Now, having experience in building software and caring of what I create and why I can do this far more quickly with LLMs and it kinda opens possibilities I could only dream of before. Like get a few spare $$$ millions and hire a team to build something before = pay $20 to Cursor/Claude and spend a few days guiding it like if it was a team of junior outsource devs: it's painful sometimes, but if you really know what you're doing and why \u2014 it works. And no one stops you from tweaking pixels when the majority of work is done \u2014 you'll even have a will to, as opposed to writing it all by yourself and spending all your mental energy on routine stuff.
So... if people learn to use this hammer properly \u2014 I suppose the future might be brighter than the past. And also those who actually care but didn't have time to do things they're passionate about now can do things on their own."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Am I getting old, or is working with AI juniors becoming a nightmare?"}},"_tags":["comment","author_rtmx","story_47888068"],"author":"rtmx","comment_text":"TBH, I'd say we were there long before LLMs came to 'help'. Software world was in a dreadful state for a decade or so, maybe longer, I'd say \u2014 devs get powerful machines not like normal users, everyone is 'just doing their job', software tested in isolated environments so no one cares about installing a couple of their own 'background services in NodeJS' on a user machine \u2014 not a big deal, yeah? And so on...
At the same time, I see the future being brighter with the help of these coding LLMs \u2014 I personally was not building software for years, focusing on management-like work. Serious coding during 'free time' was just too heavy to lift \u2014 you need time to sleep, eat and do some IRL things too...
Now, having experience in building software and caring of what I create and why I can do this far more quickly with LLMs and it kinda opens possibilities I could only dream of before. Like get a few spare $$$ millions and hire a team to build something before = pay $20 to Cursor/Claude and spend a few days guiding it like if it was a team of junior outsource devs: it's painful sometimes, but if you really know what you're doing and why \u2014 it works. And no one stops you from tweaking pixels when the majority of work is done \u2014 you'll even have a will to, as opposed to writing it all by yourself and spending all your mental energy on routine stuff.
So... if people learn to use this hammer properly \u2014 I suppose the future might be brighter than the past. And also those who actually care but didn't have time to do things they're passionate about now can do things on their own.","created_at":"2026-04-24T16:50:12Z","created_at_i":1777049412,"objectID":"47892745","parent_id":47888690,"story_id":47888068,"story_title":"Ask HN: Am I getting old, or is working with AI juniors becoming a nightmare?","updated_at":"2026-04-24T16:59:17Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jug"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Prices are also expected to drop significantly in H2 as they move to Huawei Ascend 950 super nodes.
Yes, even compared to this low price point.
As before, the headline news with DeepSeek isn't in the benchmarks, but that they're competitive there while being gut churningly cheap for the Western AI industry."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"DeepSeek v4"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://api-docs.deepseek.com/news/news260424"}},"_tags":["comment","author_jug","story_47884971"],"author":"jug","comment_text":"Prices are also expected to drop significantly in H2 as they move to Huawei Ascend 950 super nodes.
Yes, even compared to this low price point.
As before, the headline news with DeepSeek isn't in the benchmarks, but that they're competitive there while being gut churningly cheap for the Western AI industry.","created_at":"2026-04-24T16:43:07Z","created_at_i":1777048987,"objectID":"47892629","parent_id":47886400,"story_id":47884971,"story_title":"DeepSeek v4","story_url":"https://api-docs.deepseek.com/news/news260424","updated_at":"2026-04-25T06:42:32Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nedt"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"But then at the same time you should always update because it might fix a security vulnerability. Otherwise you end up running nodejs 10 because you don't need the new stuff."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Bitwarden CLI compromised in ongoing Checkmarx supply chain campaign"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://socket.dev/blog/bitwarden-cli-compromised"}},"_tags":["comment","author_nedt","story_47876043"],"author":"nedt","children":[47890846],"comment_text":"But then at the same time you should always update because it might fix a security vulnerability. Otherwise you end up running nodejs 10 because you don't need the new stuff.","created_at":"2026-04-24T11:52:16Z","created_at_i":1777031536,"objectID":"47888955","parent_id":47888811,"story_id":47876043,"story_title":"Bitwarden CLI compromised in ongoing Checkmarx supply chain campaign","story_url":"https://socket.dev/blog/bitwarden-cli-compromised","updated_at":"2026-04-24T14:33:02Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"brewtide"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"I literally just put the meshtastic antenna on the roof today, in an old services box. Been in the window for months, had a few weird perfect weather moments show a few nodes and a ping. Put it on the roof, hours ago, nothing yet.
Someone has to start up the area! (I live in nowhere maine)."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"MeshCore development team splits over trademark dispute and AI-generated code"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://blog.meshcore.io/2026/04/23/the-split"}},"_tags":["comment","author_brewtide","story_47878117"],"author":"brewtide","comment_text":"I literally just put the meshtastic antenna on the roof today, in an old services box. Been in the window for months, had a few weird perfect weather moments show a few nodes and a ping. Put it on the roof, hours ago, nothing yet.
Someone has to start up the area! (I live in nowhere maine).","created_at":"2026-04-24T01:52:45Z","created_at_i":1776995565,"objectID":"47884582","parent_id":47884140,"story_id":47878117,"story_title":"MeshCore development team splits over trademark dispute and AI-generated code","story_url":"https://blog.meshcore.io/2026/04/23/the-split","updated_at":"2026-04-25T10:29:48Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zbentley"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"There's more process-based concurrency than you'd expect in shops that use those languages.
Cron jobs might need to coordinate with webservers. Even heavily threaded webservers might have some subprocesses/forking to manage connection pools and hot reloads and whatnot. Suid programs are process-separated from non-suid programs. Plenty of places are in the "permanent middle" of a migration from e.g. Java 7 to Java 11 and migrate by splitting traffic to multiple copies of the same app running on different versions of the runtime.
If you're heavily using SQLite for your DB already, you probably are reluctant to replace those situations with multiple servers coordinating around a central DB.
Nit:
> languages that only have process based concurrency python/JS/TS/ruby
Not true. There are tons and tons of threaded Python web frameworks/server harnesses, and there were even before GIL-removal efforts started. Just because gunicorn/multiprocessing are popular doesn't mean there aren't loads of huge deployments running threads (and not suffering for it much, because most web stacks are IO bound). Ruby's similar, though threads are less heavily-used than in Python. JS/TS as well: https://nodejs.org/api/worker_threads.html"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Honker \u2013 Postgres NOTIFY/LISTEN Semantics for SQLite"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/russellromney/honker"}},"_tags":["comment","author_zbentley","story_47874647"],"author":"zbentley","comment_text":"There's more process-based concurrency than you'd expect in shops that use those languages.
Cron jobs might need to coordinate with webservers. Even heavily threaded webservers might have some subprocesses/forking to manage connection pools and hot reloads and whatnot. Suid programs are process-separated from non-suid programs. Plenty of places are in the "permanent middle" of a migration from e.g. Java 7 to Java 11 and migrate by splitting traffic to multiple copies of the same app running on different versions of the runtime.
If you're heavily using SQLite for your DB already, you probably are reluctant to replace those situations with multiple servers coordinating around a central DB.
Nit:
> languages that only have process based concurrency python/JS/TS/ruby
Not true. There are tons and tons of threaded Python web frameworks/server harnesses, and there were even before GIL-removal efforts started. Just because gunicorn/multiprocessing are popular doesn't mean there aren't loads of huge deployments running threads (and not suffering for it much, because most web stacks are IO bound). Ruby's similar, though threads are less heavily-used than in Python. JS/TS as well: https://nodejs.org/api/worker_threads.html","created_at":"2026-04-24T01:01:09Z","created_at_i":1776992469,"objectID":"47884262","parent_id":47875381,"story_id":47874647,"story_title":"Show HN: Honker \u2013 Postgres NOTIFY/LISTEN Semantics for SQLite","story_url":"https://github.com/russellromney/honker","updated_at":"2026-04-24T19:57:01Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"randerson"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Perhaps there are others in your neighborhood in the same position, who would only get into it if there were other nodes. So be the first, get your friends into it, and maybe more nodes will follow. It's only $30 or so for a device.
They have a decent range (15 miles or more) so depending on how rural you are, you might be able to create a line of repeaters back to a major population center."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"MeshCore development team splits over trademark dispute and AI-generated code"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://blog.meshcore.io/2026/04/23/the-split"}},"_tags":["comment","author_randerson","story_47878117"],"author":"randerson","children":[47884140],"comment_text":"Perhaps there are others in your neighborhood in the same position, who would only get into it if there were other nodes. So be the first, get your friends into it, and maybe more nodes will follow. It's only $30 or so for a device.
They have a decent range (15 miles or more) so depending on how rural you are, you might be able to create a line of repeaters back to a major population center.","created_at":"2026-04-24T00:24:14Z","created_at_i":1776990254,"objectID":"47884005","parent_id":47883525,"story_id":47878117,"story_title":"MeshCore development team splits over trademark dispute and AI-generated code","story_url":"https://blog.meshcore.io/2026/04/23/the-split","updated_at":"2026-04-25T10:29:48Z"},{"_highlightResult":{"author":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"nodesocket"},"title":{"matchLevel":"none","matchedWords":[],"value":"Intel's soars 15% as results top estimates"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.cnbc.com/2026/04/23/intel-intc-q1-2026-earnings-report.html"}},"_tags":["story","author_nodesocket","story_47881433"],"author":"nodesocket","children":[47881556],"created_at":"2026-04-23T20:28:37Z","created_at_i":1776976117,"num_comments":5,"objectID":"47881433","points":6,"story_id":47881433,"title":"Intel's soars 15% as results top estimates","updated_at":"2026-04-26T06:44:06Z","url":"https://www.cnbc.com/2026/04/23/intel-intc-q1-2026-earnings-report.html"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"randerson"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["nodejs"],"value":"Its optional but it helps to see where nodes are on a map, and would be useful in (for example) a search & rescue operation."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"MeshCore development team splits over trademark dispute and AI-generated code"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://blog.meshcore.io/2026/04/23/the-split"}},"_tags":["comment","author_randerson","story_47878117"],"author":"randerson","children":[47883527],"comment_text":"Its optional but it helps to see where nodes are on a map, and would be useful in (for example) a search & rescue operation.","created_at":"2026-04-23T20:00:48Z","created_at_i":1776974448,"objectID":"47880972","parent_id":47880324,"story_id":47878117,"story_title":"MeshCore development team splits over trademark dispute and AI-generated code","story_url":"https://blog.meshcore.io/2026/04/23/the-split","updated_at":"2026-04-25T11:57:33Z"}],"hitsPerPage":20,"nbHits":31213,"nbPages":50,"page":0,"params":"query=nodejs&advancedSyntax=true&analyticsTags=backend","processingTimeMS":9,"processingTimingsMS":{"_request":{"roundTrip":13},"afterFetch":{"format":{"highlighting":1,"total":1}},"fetch":{"query":2,"scanning":5,"total":8},"total":9},"query":"nodejs","serverTimeMS":11}