Claude mixes up who said what
Claude sometimes sends messages to itself and then thinks those messages come from the user. This is categorically distinct from hallucinations or missing permissions.
The latest blog posts to take off on Hacker News. 1,521 posts.
Claude sometimes sends messages to itself and then thinks those messages come from the user. This is categorically distinct from hallucinations or missing permissions.
With Anthropic altering rate limits, we can find other places to spend our money. I have chosen Zed editor with OpenRouter for maximum flexibility.
In 1958, Mao ordered every village to produce steel. The steel was useless. The crops rotted. Today's top-down AI mandates are producing the same pattern: ba...
Update (April 9, 2025): Since publishing this post, a human support agent from Anthropic has reached out to me. I appreciate them following up.
Claude Managed Agents is Anthropic's suite of composable APIs for building and deploying cloud-hosted agents at scale. See what's new and how to get started.
Today, we’re excited to introduce Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs. Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.
This is a long article, so I've broken it up into a series of posts, listed below. You can also read the full work as a PDF or EPUB .
Technical details on how we tested Claude Mythos Preview, a model strikingly capable at computer security tasks, and what we found.
A new initiative to secure the world’s most critical software and give defenders a durable advantage in the coming AI-driven era of cybersecurity.
KL-divergence is a topic which crops up in a ton of different places in information theory and machine learning, so it's important to understand well. Unfortunately, it has some properties which seem confusing at a first pass (e.g. it isn't symmetric like we would expect from most distance measures, and it can be…
AI made competent output cheap. That raises the value of taste and exposes its limits. The real edge is judgment paired with context, stakes, and the will to build.
Large language models may be standardizing human expression and subtly influencing how we think, says study led by USC Dornsife researcher
Anthropic secures multiple gigawatts of next-generation TPU capacity starting in 2027, its largest compute commitment to date.
Bad software is a choice you make
Over the past two years I’ve quietly rebuilt major parts of the OldNYC photo viewer . The result: 10,000 additional historic photos on the map , more accurate locations, and a site that’s cheaper and easier to run—thanks to modern AI tools and the OpenStreetMap ecosystem.
As the eleventh hour dawns all the pieces start to fall into place. I lived my life knowing this would happen, yet when it is I may be just as unprepared as anyone else. As any self driving car maker knows, predicting doesn’t mean you can act.
Web developer and cybersecurity researcher
LM Studio 0.4.0 introduced llmster and the lms CLI. Here is how I set up Gemma 4 26B for local inference on macOS that can be used with Claude Code.
For eight years, I’ve wanted a high-quality set of devtools for working with SQLite. Given how important SQLite is to the industry1, I’ve long been puzzled that no one has invested in building a …
Update (9 Apr 2026): Another one. A helpful person sent in Microsoft 365 Copilot for US Government. That's 81. Chart updated.
How coding agents use tools, memory, and repo context to make LLMs work better in practice
All modern language models sometimes act like they have emotions. What’s behind these behaviors? Our interpretability team investigates.
We replaced expensive sandboxes with ChromaFs, a virtual filesystem over Chroma, to give our docs AI assistant the ability to explore documentation like a developer would.
Cursor 3 is a unified workspace for building software with agents.
Our most intelligent open models, built from Gemini 3 research.
Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
What the accidental Claude Code source code leak tells us about the real value of code, product market fit, and why integration is what actually makes software great.
You may already be aware of John Baez’s “Crackpot Index” , which provides a simple tongue-in-cheek scoring system for assessing whether a research contribution to physics is likely “revolutionary.” The list contain such gems as “40 points for comparing yourself to Galileo, suggesting that a modern-day Inquisition is…
Meta is continuing its long-term roadmap to help the construction industry leverage AI to produce high-quality and more sustainable concrete mixes, as well as those exclusively produced in the Unit…