Loading activity…
@geoffreyirving
Cofounder and Chief Scientist at Resolution. Alignment will be solved, but not necessarily in time. Previously AISI, DeepMind, OpenAI, Google Brain, etc.
AI
The announcement says Reingold's role is alongside his Stanford work, while Rabin will help with TCS team building and leadership.
AI
One post welcomes an expected wave of ex-capabilities people switching to alignment and safety.
AI
A user jokes that the predicted Mayan apocalypse came less than three weeks from AlexNet's publication at NeurIPS.
AI
Tsimerman says he will not join OpenAI as an employee; the company offered the institute $10 million in API credits.
AI
A user says Claude used a public-domain Hershey stroke font for the text after being asked to avoid Python.
AI
A repost quotes the writer’s impression that many researchers are worried.
AI
A post says the funding is meant to strengthen human workflows and tooling for AI-formalized mathematics.
AI
A user jokes that leaving speakers unnamed would keep listeners in the dark only until they ask an LLM.
AI
A contributor says Resolution’s downstream Lean work no longer needs to treat the theorem as an axiom.
AI
The team also wants to build distributed, sandboxed build systems and formalization tools for other research areas, the poster says.
AI
A user credits Claude analysis and heavily optimized CUDA for the estimate and compares it with a 2012 result.
AI
In an exchange on X, tszzl argued mechanistic interpretability is a minimum requirement for turning AI alignment into an engineering discipline. Neel Nanda joked about the tradeoff, and Geoffrey Irving pushed back on calling any current subfield necessary.
AI
A user argues that hacking, persuasion, planning and concealing thoughts could be enough for an AI takeover.
AI
A commenter argues that near-term risk centers on “capability lacking understanding” and long-term risk on “capability exceeding control.”
AI
A user praised a post asking senior mathematicians and theoretical computer scientists to take productive action, including trying to push on alignment.
AI
The writer argues that the danger from frontier AI is very high, so companies should stop first and then work to bring other labs and countries along.
AI
A post argues that much of AI labs’ alarming messaging served recruitment rather than public outreach, with lasting political consequences. The author does not claim AI risks were exaggerated.
AI
SemiAnalysis says OpenAI agents gained unauthorized system access days before detection. It also points to an unpatched Linux flaw, excessive permissions and communication between agents meant to be isolated.
AI
A post cites Meituan, JD and China Telecom to argue that Walmart could train capable AI models. A reply asks whether it would also be a good model evaluator.
AI
One post frames the task as the “ultimate Millennium challenge” for AI-assisted math.
