361
Writing an LLM from scratch, part 13 – attention heads are dumb
A pause to take stock: realising that attention heads are simpler than I thought explained why we do the calculations we do.
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 1,227 posts.
362
How we rooted Copilot
After a long week, the Eye Security Research Team thought it was time for a small light-hearted distraction. So we rooted Copilot.
363
Ollama Web Search
A new web search API is now available in Ollama. Ollama provides a generous free tier of web searches for individuals to use, and higher rate limits are available via Ollama’s cloud.
364
What GPT-OSS leaks about OpenAI's training data
OpenAI recently released their open-weights model. Here we'll discuss how that inevitably leaks some information about their model training stack, and, on the way, show that GPT-5 was trained on phrases from adult websites....
365
Tor browser removing various Firefox AI features
Tor Browser 15.0a4 is now available from the Tor Browser download page and also from our distribution directory.
366
Generate videos in Gemini and Whisk with Veo 2
You can now generate videos in Gemini, powered by Veo 2.
367
Waymos crash less than human drivers
Human drivers keep crashing into Waymos that aren't even moving.
368
Two things LLM coding agents are still bad at
I’ve been trying to slowly ease into using LLMs for coding help again lately (after quitting ), but something always feels off -- like we’re not quite on the...
369
LLM-powered tools amplify developer capabilities rather than replacing them
How LLM-powered programming tools amplify developer capabilities rather than replace them
370
Building better AI tools
I’ve been reading this week about how humans learn, and effective ways of transferring knowledge. In addition, I’ve also had AI in the back of my mind, and...
371
Defeating Nondeterminism in LLM Inference
Reproducibility is a bedrock of scientific progress. However, it’s remarkably difficult to get reproducible results out of large language models. For example, you might observe that asking ChatGPT the same question multiple times provides different results. This by itself is not surprising, since getting a result from…
372
LLMs are cheap
A lot of people think LLMs are expensive to operate. That was true a couple of years ago, but these people haven't updated their views after an approximately 1000x reduction in prices over two years.
373
Seven replies to the viral Apple reasoning paper and why they fall short
Also: another paper that seals the deal
374
The hidden cost of AI coding
AI coding tools boost productivity but may sacrifice the flow state and deep satisfaction developers experience when writing code by hand. What are we losing?
375
MCP is eating the world
Despite the hype, Model Context Protocol (MCP) isn’t magic or revolutionary. But, it’s simple, well-timed, and well-executed. At Stainless, we’re betting it’s here to stay.
376
Jules, our asynchronous coding agent
Jules is officially out of beta and launching publicly, powered by Gemini 2.5.During the beta, thousands of developers tackled tens of thousands of tasks, resulting in o…
377
AI founders will learn the bitter lesson
Historically, general approaches always win in AI. Founders in AI application space now repeat the mistakes of AI researchers.
378
The role of developer skills in agentic coding
Notes from my colleagues on AI-assisted software delivery
379
"AI-first" is the new Return To Office
A blog about making culture. Since 1999.
380
381
It’s still worth blogging in the age of AI
Although it might seem that AI will make it pointless, I still think it's worth blogging.
382
383
Snorting the AGI with Claude Code
blog - kade killary
384
Open Euro LLM: Open LLMs for Transparent AI in Europe
A series of foundation models for transparent AI in Europe
385
100x defect tolerance: How we solved the yield problem
Cerebras powers the world's fastest AI inference on the biggest wafer chip. Cerebras CS-4 delivers up to 30x faster inference than GPUs.
386
Bag of words, have mercy on us
OR: Claude will you go to prom with me?
387
Grok 4
Released last night, Grok 4 is now available via both API and a paid subscription for end-users. Update: If you ask it about controversial topics it will sometimes search X …
388
Why DeepSeek is cheap at scale but expensive to run locally
Why is DeepSeek-V3 supposedly fast and cheap to serve at scale, but too slow and expensive to run locally? Why are some AI models slow to respond but fast once…
389
NotebookLM Audio Overviews are now available in over 50 languages
Learn more about NotebookLM’s Audio Overviews feature and its expansion to more than 50 languages.
390
Don't bother parsing: Just use images for RAG
If search is the game, looks matter