Loading activity…
@mmitchell_ai
Interdisciplinary researcher focused on shaping AI towards long-term positive goals. ML & Ethics. Similar content in the Skies (this bird has flown).
AI
The chatbot spelled 17 “seveen,” one commenter says; another says reasoning was off for speed and comedy.
AI
Google is making its AI tools easier to restrict, but schools blocking AI entirely will lose students' Search access.
AI
The reported MATH-500 score rose from 7.9% to 48.1% with the same sampling settings.
AI
On X, @chrisalbon described an AI dinner suggestion; replies proposed asking for novel ideas or restructuring examples.
AI
IEEE Spectrum describes researchers’ disagreement over what to tell AI models about their own identities.
AI
A user cites Lisanne Bainbridge's 1983 paper to argue that AI oversight can overwhelm humans and erode their skills.
AI
Verena Rieser says her Budapest keynote will examine the AI frontier’s interest in constitutions and NLP’s role.
AI
A Tech Policy Press perspective argues safety assessments should test model behavior, not just review developer-written system prompts.
AI
Mitchell and her coauthors argue that risks to people rise as users cede more control to AI agents.
AI
One commenter doubts the FDIC could untangle agent traces and possibly falsified logs; another cites protections for unauthorized transfers.
AI
Two posts point to security incidents and urge AI-assisted defenses; one specifically calls for faster vulnerability detection and patching.
AI
In a reply, a user says getting the point right requires seeing the specific instructions OpenAI gave the agents.
AI
A user points to a co-written essay, an interview with Ezra Klein and writing by others with similar views.
AI
Clement Delangue says it omitted the lack of agent containment and monitoring and Hugging Face's open-model defense.
AI
A post quotes David Sacks arguing that Claude’s training could increase the risk of frontier AI escaping human control.
XAI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
One poster suggests feeding models RAM, electricity and clock-cycle signals as a form of “interoception.”
AI
A user questions whether reasoning they attribute to Anthropic assumes a model is conscious to conclude that it is.
AI
David Robinson argues OpenAI’s trial-and-error safety culture is inadequate for increasingly capable AI systems.
AI
One post claims OpenAI paused tool-use training, evaluations and inference on its most capable models while it fixes containment.
