A Claude model called Fable reportedly produced a counterexample to the Jacobian Conjecture, one of math's longest-standing open problems. Meanwhile, Moonshot AI suspended new Kimi K3 subscriptions due to overwhelming demand.
A model called Claude Fable may have produced a counterexample to the Jacobian Conjecture — a problem that has stumped mathematicians since 1939. If verified, this is not a benchmark score or a coding demo. It is a genuine mathematical result. Builders should care because it signals that frontier AI is moving from 'useful assistant' to 'active research collaborator'.
The Jacobian Conjecture asks whether a certain class of polynomial maps must always be invertible. It has resisted proof for over 80 years and sits on most mathematicians' lists of hardest open problems. A post on X from researcher Alpoge claims Claude Fable found a counterexample — meaning the conjecture is false, not just unproven. The claim is circulating with 502 Hacker News points as of today. It has not been peer-reviewed. Treat it as a very exciting lead, not a confirmed result.
Moonshot AI has suspended new subscriptions to its Kimi service because demand for Kimi K3 is outpacing capacity. The announcement came via the official Kimi Moonshot account on X. If you already have a subscription, you are fine. If you were planning to sign up to test K3, you are currently locked out. No timeline was given for when signups reopen.
A study covered by The Next Web found that people who received AI advice on a task got more answers wrong — but felt more confident in those wrong answers. This is the 'automation bias' problem in a new coat. For builders shipping AI-assisted tools: if your product gives recommendations, users may stop double-checking. That is a design problem, not just a research footnote.
Baidu's Unlimited-OCR model hit the #4 spot on Hugging Face model rankings, cross-confirmed by both the models and spaces leaderboards. If you are building anything that needs to pull text from images, PDFs, or scanned documents, this is worth a look. It is available directly on Hugging Face.
Cline, the AI coding agent that runs inside VS Code, shipped v4.0.10. llama.cpp hit build b10069. Both are incremental releases — no headline features announced — but if you run local models or use Cline daily, pull the updates.
1. Test Baidu's Unlimited-OCR on a real document extraction task. Drop a messy PDF or a photo of a receipt into the Hugging Face Space and see if it beats whatever you are currently using. No code required.
2. If you ship an AI recommendation feature, run a quick user test this week specifically looking for automation bias. Ask users to explain why they accepted a suggestion. If they cannot, your UI may need a 'show your reasoning' step.
3. Watch the Jacobian Conjecture thread closely. If the counterexample holds up under scrutiny, it will be the clearest public proof yet that frontier models can do original research — and that changes how you should think about using them for hard domain problems, not just drafting and summarizing.
Confirmed: Moonshot AI officially announced suspension of new Kimi K3 subscriptions via its verified X account.
Not independently verified by CBW: The Claude Fable Jacobian Conjecture counterexample has not been peer-reviewed. The claim comes from a single X post. We have not seen the actual mathematical output.
Not independently verified by CBW: The study on AI advice and confidence is reported via The Next Web. CBW has not read the underlying paper or checked its methodology.
Confirmed: Baidu's Unlimited-OCR is live on Hugging Face and cross-confirmed across both models and spaces rankings.
Worth noting: 'Claude Fable' is not a publicly documented Anthropic model name as of this writing. It may be an internal or research variant. Anthropic has not issued a press release about this result.
Source: X — Alpoge on Claude Fable and Jacobian Conjecture — https://xcancel.com/__alpoge__/status/2079028340955197566
Source: X — Kimi Moonshot subscription suspension announcement — https://twitter.com/kimi_moonshot/status/2078855608565207130
Source: The Next Web — AI advice suppresses critical thinking study — https://thenextweb.com/news/ai-advice-suppresses-critical-thinking-wrong-answers-study
Source: Hugging Face — Baidu Unlimited-OCR model — https://huggingface.co/baidu/Unlimited-OCR
Source: GitHub — Cline v4.0.10 release — https://github.com/cline/cline
Source: GitHub — llama.cpp build b10069 — https://github.com/ggml-org/llama.cpp
Get tomorrow's best AI project in your email. With a guide that works. Free. No spam.