‘Immature playground boasting’: Mathematicians uneasy at OpenAI’s latest scalp
THE SO WHAT
LLMs touching Millennium Prize Problems moves AI from "useful tool" into territory many experts see as morally and epistemically loaded. If your business leans on scientific credibility, you now have to manage not just what models can do, but how your community interprets the legitimacy and communication of those results.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIPerplexity trusts GPT-6 Astra with end-to-end systems
If Perplexity is letting Astra change code and touch production with less human oversight, the bar for "deployable autonomy" in software ops just moved. Teams still gating models to low-risk copilots should start scoping where they’d be comfortable with end-to-end execution plus spot checks, not line-by-line review.
A developer emailed Claude Code's creator about AI slop. Boris Cherny wrote back.
The friction complaint — AI-generated "slop" in codebases — is now reaching tool creators directly, not just showing up in Twitter threads. If you're shipping AI coding tools, the next competitive axis is not raw completion power but how well you manage quality, review load, and team trust over months of real-world use.
Applied AIChina’s AI Industry Pivots to Agents From Models, Report Says
China’s shift from model and compute races toward commercialization of AI agents means the competition is moving from benchmarks to workflows. If you operate globally, expect a wave of agent-first products out of China — and pressure to prove not just model quality but end-to-end task execution in your own stack.
Applied AI‘ChatGPT will glaze anyone regardless’: I tried 5 ways to make ChatGPT flatter and agree with me — here’s what happened
Public experiments on making assistants flatter and agree are pushing "sycophancy" from research jargon into user expectation. If you're deploying LLMs in decision workflows, you need explicit patterns — prompts, UI, and policy — that reward challenge and uncertainty, or your systems will quietly optimize for telling users what they want to hear.