@tszzl
“ceterum censeo we must pace the frontier of global machine intelligence progress”
AI
One user puts Hindutva, leftist and film Twitter atop the list, then adds open-model, “4o” and VC Twitter.
AI
One commenter says it's hard to separate distillation's contribution from gains made through internal reinforcement learning.
AI
The poster says surveyed Anthropic staff's answers imply median annual US GDP growth of 17.1% in 2030 under Anthropic's model.
AI
In a reply, a user says AI alignment hasn’t been solved and urges another user not to say otherwise.
AI
Anthropic says it is in discussions with multiple organizations about third-party evaluations of its safety practices.
AI
Goodfire researcher Nick Cammarata argues that modern tools reveal most neurons as legible in models up to trillions of parameters.
AI
A post says hills are rare in high-dimensional space, where you “end up dying on saddle points.”
AI
A post likens the labs' marketing to a priest shouting the tetragrammaton at a pitch meeting.
AI
A user suggests that perhaps only xAI among major AI labs is not technically headquartered in San Francisco.
AI
A post says Jack Clark moved up his RSI timeline and described creativity as the last piece.
AI
A user points to models getting good enough to do research and encouraging developments they say aren't theirs to share.
AI
A user recalls GPT-4 getting confused by elementary school story problems in 2023.
AI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
A user argues that mechanistic interpretability is a minimum requirement for making AI alignment an engineering discipline.
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 asks whether the protections have helped anyone contact California’s attorney general.
AI
A post links the term to slightly misspecified objective functions.
AI
Clark shared a remark heard at The Curve and imagined future thieves, security systems and policing.
AI
Elon Musk called spreading superintelligence to the stars a “great success condition”; a reply urged him not to give up on humans.
