Loading activity…
@sauers_
gnome
AI
A user asks how models could learn to do this and points to a thread.
AI
A user says Opus won't recognize what appears to be its latest message when asked if it said it.
AI
A user links to a photo but does not explain how the steering works.
AI
A user says replacing the 'gate' component with a constant did not significantly degrade the model's behavior.
AI
A user calls the claimed milestone “huge” in a quote post linking to Apollo Research.
AI
A repost quotes a speaker rejecting a 'compiler hijack' to avoid becoming a famous example of deceptive AI.
AI
A repost relays someone’s uncertain recollection that Anthropic posted a graphic while discussing how misalignment scales.
AI
A poster says almost any Stage 4 cancer patient would find human understanding of a working cure's mechanism irrelevant.
AI
The researchers say a Qwen student's letter-counting accuracy rose from 27.3% to 59.8% after training on unrelated answers.
AI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
A user jokes that Claude Opus 4 “came back from the dead,” marking the post as sarcastic with “/s.”
AI
A user shared a photo link with only “virtual structures” as a caption.
Technology
Human lab work followed a 21-hour search by roughly 950 Claude agents; the system's function and possible uses remain unknown.
AI
One user argues that recent GPT models hedge too much in biological reasoning, leading them to remove reasonable conclusions.
AI
A commenter argues that liability and tort law should apply to safety evaluators, too, if they ask models to do dangerous things.
AI
CAIS argues that the AI safety label covers two “radically different projects.”
AI
One post calls the code “terrific,” highlighting its emphasis on augmenting human roles rather than competing with them.
AI
Stolen Thoughts claims encrypted reasoning blocks from Anthropic, OpenAI and Google APIs are interchangeable across sessions, users and models—and says it uses this to decode hidden reasoning at scale.
AI
The critic says Anthropic has published little about its pre-retirement interviews with AI models and argues that simply talking to them is no meaningful welfare commitment.
