Where the AI market is talking

Fermat's Last Theorem in Lean 4

I don't understand what you are trying to say. Which of the following is it, (or is it something else entirely)? 1. There is a much shorter proof that would also be accepted by lean, we just aren't thinking about the problems in the right way so we can't find it. On one level this is obviously true, Anthropic did not put any effort in to minimising the length of the proof during its development or afterwards. 2. There is a much shorter proof if we took different axioms instead of the ones built into lean. I find this much harder to believe, unless your new axiom is basically just FLT. Otherwise all reasonable axioms are not too hard to show as equivalent to each other (in terms of what they prove in PA anyway), so such an equivalence proof would be a small portion of the 13 million lines …

Claude's new system prompt doesn't want to reproduce song lyrics

A few days ago I saw some people criticizing Dwarkesh’s explanation of the HuggingFace hack by OpenAI’s AI agents. Their point was that by anthropomorphizing the AI agents, accountability of and blame on OpenAI’s poor practices are being ignored. Now I see this blog post and wonder if Anthropic is being more true to its name and moving to “AI sentience” with the below. > The way they handle abusive conversations has changed a bit too. The previous Fable 5 system prompt included this: > If the person becomes abusive or unkind to Claude over the course of a conversation, Claude maintains a polite tone and can use the end_conversation tool when being mistreated. Claude should give the person a single warning before ending the conversation. “Mistreated”? Can GenAI be mistreated? It’s ju…

Claude's new system prompt doesn't want to reproduce song lyrics

A few days ago I saw some people criticizing Dwarkesh’s explanation of the HuggingFace hack by OpenAI’s AI agents. Their point was that by anthropomorphizing the AI agents, accountability of and blame on OpenAI’s poor practices are being ignored. Now I see this blog post and wonder if Anthropic is being more true to its name and moving to “AI sentience” with the below. > The way they handle abusive conversations has changed a bit too. The previous Fable 5 system prompt included this: > If the person becomes abusive or unkind to Claude over the course of a conversation, Claude maintains a polite tone and can use the end_conversation tool when being mistreated. Claude should give the person a single warning before ending the conversation. “Mistreated”? Can GenAI be mistreated? It’s ju…

Formalizing Fermat's Last Theorem

You can say that American society made OpenAI and Anthropic possible. No other current society would have. Suddenly, formalisation of math is becoming cheap. That's not a problem, that's the goal, and it is here much earlier than expected. That's not antisocial. That is scientific progress. (I swear, did not use an LLM for this)

Claude's new system prompt doesn't want to reproduce song lyrics

I really wish Anthropic would publish those tool descriptions. I learned a ton of useful things about ChatGPT Work by having it dump out its tool descriptions the other day: https://codex-tool-reference.simonw.chatgpt.site/

Corporate America is getting hooked on open-source AI

... and practically nobody's got the money for Fable at API prices. with GLM 5.3 Flash being a Reasonable Opus 4.x Substitute, Anthropic's moat is effectively gone.

Poetry book that Anthropic tried to censor

The post's explanation for the content filtering policy 400 error isn't correct. Anthropic doesn't use that error for safety reasons ("These refusals do not reflect Anthropic’s judgments about the propriety of any content"[1]); content blocked by classifiers get a different kind of error. The content filtering 400 is only for copyright blocks, where Anthropic detects when Claude is outputting copyrighted text and blocks it to prevent copyright infringement. The book in question is in the public domain in the US (but not all countries) as of January 1, 2026; Anthropic probably just doesn't automatically remove works from the copyright filter when their copyright expires. [1]: https://privacy.claude.com/en/articles/10023638-why-am-i-rec...

Claude's new system prompt doesn't want to reproduce song lyrics

With the lyrics in modern pop songs, I don’t blame Claude. I guess Anthropic really does care about his well-being.

Poetry book that Anthropic tried to censor

Claude's new system prompt doesn't want to reproduce song lyrics

I like Simon but the behavior itself isn’t new. Just make sure claude understands you’re not needing Claude to output the lyrics as you both have them. Which also lessens the issue with accidental sharing. It’s not as shocking nor concerning if you start thinking about claude like a contractor that works for you through Anthropic. Anthropic has rules for their employees. Like any contracting arrangement, collaboration finds a way.

Claude's new system prompt doesn't want to reproduce song lyrics

It’s curious that while there is talk of certain SOTA models being on the brink of AGI, Anthropic doesn’t word this part of the system prompt in terms of copyright and plagiarism, such that Claude would be able to judge on its own which reproductions are appropriate or not. As long as we’re seeing things like that, it’s saying a lot about the AI companies’ trust in the capabilities and reliability of their models and harnesses.

Ask HN: Resources to get good at soldering?

My skill went up quite a bit when I started watching Louis Rossmann's explainer videos. He mainly does component level repair of macbooks and he covers his troubleshooting approach and the equipment he uses is nothing you can't replicate at home. I don't bother pulling out microscope for repairs and just stick with magnifying glass.

Microsoft's MAI-Transcribe-2 Undercuts OpenAI, Google and ElevenLabs

Ask HN: Who wants to be hired? (September 2026)

Location: Istanbul, Türkiye Remote: Yes Willing to relocate: No Timezone: UTC+3 Technologies: TypeScript, JavaScript, Next.js, React, Node.js, Python, FastAPI, PostgreSQL, Supabase, MongoDB, OpenAI APIs, LangChain, n8n, Vercel, ComfyUI Full-stack developer focused on AI products, agentic systems, and automation. My recent work has been around building practical LLM-powered systems rather than simple chat interfaces: multi-step agent workflows, tool/function calling, structured outputs, persistence, approval pause/resume flows, retries/checkpoints, streaming, API integrations, and automation pipelines. Projects include PersonaChat, AI-powered web products, Discord bots, ComfyUI/ComfyDeploy workflows, and agent systems combining backend logic, databases, external APIs, a…

Phaseone[big] explains why you should trust OpenAI

had a lot of fun putting this Pluribus edit together. in the original scene, the hive explains to carol how they eat dead humans made with gpt-6 astra, elevenlabs for the voice, and a lip sync model from fal astra chose to just use ffmpeg directly for the video editing. i then manually added the captions using descript

Show HN: RagLeap Core – 46 AI Employees, open-source LangChain alt

Thanks for checking! Built this as open alt to LangChain/CrewAI - 46 role-based AI Employees that run from your own docs on your own server. Self-hosted, BYOK, MIT. Stack is 4 PyPI packages (26k downloads) - ragleap-rag with 6 vector DBs + hybrid search, ragleap-graph with Neo4j. Currently at 0 stars, day 0 - Would love feedback from anyone who built RAG in production - what broke for you?

Show HN: Mu, two-minute visualized lessons on modern AI

Hi HN, I am Sreedath Panat. A year ago, I had started a habit of making small explainer videos of 2-3 minutes to help me understand, in a nutshell, some of the difficult-to-understand research papers. I'm not saying that this gave me a deep understanding, but it definitely helped me set foot into some of the research papers (eg: I-JEPA, flamingo paper etc). Then I thought, why can't the same technique be used for getting a minimal understanding of different concepts within AI? There is a reason why I try to convert this thought into a simple product. The 2-minute micro lessons are fully built using extensive workflow I have set up in Claude Code with a good voice agent from ElevenLabs. I have extensively tested this myself. If I find the micro videos interesting or not, I found it extreme…

Show HN: RagLeap Core – 46 AI Employees, open-source LangChain alt

Ask HN: Replicate ChatGPT/Anthropic Voice Mode?

Hi, I'm interested in how to have my own phone number where I can call in and get a similar experience to ChatGPT voice mode or Anthropic voice mode. Anyone have experience setting this up? I've got something rudimentary setup with Parakeet for STT, Elevenlabs for TTS, Claude for the brains, and Twilio for the phone number handling. It would have impressed someone in 2024 but...

“Next-token predictor” is the wrong mental model for LLMs

> It's just not predicting based on it's training data, but predicting based on RLVR & more, trying to get to the optimal solution ( as much as the solutions CAN be optimal) It is predicting based on a model. In many cases we can download the model off hugging face. The model is conditioned by all sorts of things. Training data, post-training, coincidence, prompt inputs, runtime data available from whatever means. > but at least I would still call it a "next token predictor" We can call any prediction system a next token predictor. If you watch over the shoulder of a human writing a HN comment you are almost certain to see them generating a linear string of tokens. That is what keyboards do. It is impossible to generate text without being equivalent to a next token predictor.