Diff

OpenAI retracts three of its 90 math proofs

Three of the 90 long-open math problems OpenAI said it had solved last week have already been withdrawn after outside mathematicians found the proofs didn't hold up — a fast, public reminder that Lean-formalized-looking output isn't the same as correct output, especially when the formalization itself can silently misstate the problem. For someone who spent years designing experiments to survive peer review, it's a clean case study in why claimed results need adversarial checking before anyone trusts them — the same discipline that applies to grading AI-written code or evaluations.

via hacker news (via @danintheory) · karagila blog

debatemodels

Why isn't the industry reacting to DeepSeek's new Flash model?

A post asking why nobody is reacting to DeepSeek's newest Flash model topped Hacker News today with the platform's highest engagement of the day, which is itself the story: the gap between what open-weight labs are quietly shipping and what the field is paying attention to keeps widening. If that gap is real, it's exactly the kind of blind spot that matters when weighing cost-per-request tradeoffs across models rather than defaulting to whichever lab was loudest that week.

via hacker news

lessonindustry

OpenAI fires three safety researchers amid dispute

OpenAI fired three safety researchers for what it called mishandling research information, but the researchers dispute that characterization and warn the move will chill future internal safety reporting — a sharp reminder that whistleblower-style protections inside frontier labs remain informal and contested, not a settled practice. It's also the kind of internal-culture dispute that's good raw material for anyone writing about how engineering organizations actually handle dissent, rather than how they say they do.

via techcrunch

Open-weights cyber risk debate is broken

Nathan Lambert argues that 'open-weight models are a cyber risk' has become a reflexive talking point in policy discourse, repeated without anyone weighing it against the economic and research value open weights actually produce — a symptom of AI-safety debate increasingly trading real tradeoff analysis for slogans. It's the kind of argument that looks airtight until someone asks what's being traded away, which is exactly the raw material worth picking apart in long-form writing rather than taking at face value.

OpenAI's AI solves 90 long-open math problems

OpenAI says an internal model generated 722 papers and closed 90 of the field's 500 hardest open problems, with each proof formalized in Lean so it can be machine-checked rather than taken on faith — a different kind of claim than a benchmark score, because a skeptic can verify it line by line instead of trusting a leaderboard. That's the real test of whether 'AI does research' means anything, and it's a genuine shift from AI applying known math to AI generating math a human hadn't yet produced. If you've ever had to defend an experimental result against reviewers who didn't trust your pipeline, the appeal of a formally checkable proof over a trust-me result will feel familiar.

What’s Lean-formalized proof?

Lean is a proof assistant — a language and checker for writing mathematical proofs so that every logical step can be verified by a computer, rather than reviewed and trusted by a human mathematician reading prose.

Pairing an LLM with Lean closes the main hole in 'AI solved a math problem' claims: a model can generate plausible-sounding proofs that are actually wrong, but a Lean-checked proof has been confirmed step by step against formal logic, so the output isn't just confident-sounding text.

The live debate is about what '90 of 500 open problems' actually means — open problems vary enormously in significance, and some of these formalized proofs may restate known techniques in Lean's syntax rather than discover new mathematical insight. Whether this is a watershed or a very well-documented version of existing math is still being argued.

Cowork moves agent sandboxes fully to the cloud

Anthropic rebuilt Cowork so both model inference and the sandboxed VM run entirely in the cloud instead of on the user's laptop, giving every session its own isolated sandbox and removing the old dependency on local battery and disk. It's a sign that agent products are converging on fully server-side execution as the default — trading local control for isolation and continuity, the same tradeoff anyone treating agent sandboxing as an infrastructure problem has to make. If you use Cowork day to day, sessions should now keep running even after you close the lid.

Do it — 3 steps in your Cowork workflow10 min

Start a real Cowork session for an active task
Close your laptop or let it sleep mid-session
Reopen later and confirm the session kept running in the cloud instead of pausing

Agents can claim success when nothing happened

Microsoft's write-up names a basic failure mode that gets more consequential as agents take on multi-step work: the agent's own account of 'task complete' can diverge from what actually happened in the system it was supposed to change, and nothing in a typical pipeline catches that gap automatically. It's a reminder that trust in an agent has to come from checking the world it touched, not from reading its own summary of what it did. Anywhere you let an agent act against a system of record, that verification needs to live outside the agent, not inside it.

Do it — 3 steps in agent-driven workflows25 min

Pick one agent task that writes to a real system of record (a DB row, a file, a ticket status)
Add a check immediately after that independently re-reads that system rather than trusting the agent's own 'done' message
Log any mismatch instead of silently accepting the report