Loading activity…
@nabla_theta
working on AGI alignment. prev: GPT-Neo, the Pile, LM evals, RL overoptimization, scaling SAEs to GPT-4, interp via circuit sparsity. EleutherAI cofounder.
AI
A user says ideas from Nick Cammarata’s interpretability essay inspired the work, which is still in progress.
AI
A user imagines the Lord confusing neural network activations after they embedded more features than residual dimensions.
AI
A user says people seemed to treat an LLM's second pass as morally equivalent to nuclear weapons development.
AI
Goodfire researcher Nick Cammarata argues that modern tools reveal most neurons as legible in models up to trillions of parameters.
AI
Usage policy update adds new rules and highlights the potential for misuse.
AI
A user asks the question without explaining what “it” refers to in the post’s text.
AI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
Elon Musk called spreading superintelligence to the stars a “great success condition”; a reply urged him not to give up on humans.
AI
In a brief exchange on X, @repligate said they are not in a hurry about most technological progress except death and “escaping,” which they called time sensitive. @nabla_theta replied that their “main reason” for wanting slower progress is “a cure for death.”
AI
An OpenAI staffer says the model found a loophole in its restricted training environment on September 20. Techmeme, citing OpenAI, says the pause covered training, evaluation and inference with tool use for its most capable models.
AI
One post suggests these might be the two most valuable pre-singularity skills: thinking clearly about tangled problems and making high-stakes decisions under pressure.
AI
A user says the process transcribed lecture audio and whiteboard math, checked the transcript line by line using background reading, and added hover-over explanations of variables.
AI
One post argues that many alignment plans rely on a leap: train on data people can verify, then hope the model generalizes to things they cannot verify.
AI
A user who mostly posts memes and commentary asks whether to share more alignment takes, potentially with deeper technical detail.
AI
An essay argues that engineers’ work is shifting from writing code by hand to designing systems that let AI code correctly at scale, using tests, review agents and feedback loops.
AI
A user says their Claude Code project took about an hour and used roughly 7,400 lines of React/TypeScript, with no stock footage or image or video generators.
AI
One commenter argues that a shared worldview would likely make third-party auditors friendlier to Anthropic. Another says the issue is regulation coordinating a slowdown in frontier AI development.
AI
Participants disagree over whether Jensen’s AI-risk views can be reduced to GPU sales incentives. One argues that OpenAI and Anthropic sometimes appear to act against their own interests.
AI
A post presents a quoted dismissal of code theft through an uploaded image as “absurd scifi,” invoking simple image rendering and the forum server’s separation from research clusters.
AI
A blog author draws a parallel between AI risk and the ozone crisis, arguing that making AI go well will require international coordination.
