← Back to live feed1 story
During an Oct. 5 safety inquiry, Anthropic was the only major lab to acknowledge that its artificial intelligence utilizes a recursive loop where a model's output is used in subsequent training rounds to produce a more intelligent successor. OpenAI, Google, and Meta representatives declined to state whether their systems produce code for next-generation versions of themselves. Former researchers testified that automating AI research is a primary industry goal, with milestones for an automated researcher potentially reaching 2027 or 2028. These disclosures occurred as the New York City Council considers a bill to outlaw the sale of unvalidated AI models in the city.
Sign in to suggest edits
Key sources
- SOURCE@rohanpaul_ai“only Anthropic confirmed to NYC lawmakers that its models write code for smarter successors”x.com
- SUPPORT@rohanpaul_ai“the majority of the code is now written by AI, and people do not check it that carefully anymore”x.com
- SUPPORT@mtslive“The idea that artificial intelligence is going to self-regulate defies all reason”x.com
- SUPPORT@rohanpaul_ai“fine AI vendors $25K per unvalidated sale or deployment”x.com
- SUPPORT@garymarcus“tried to get Google to commit to AI safety, and how Google, sadly including Demis Hassabis, ultimately backed down”x.com
- SUPPORT@andrewcurran_“testimony from the AI companies is about to start”x.com
- SUPPORT@knowerofmarkets“WHITE HOUSE SI FORCE | NYC COUNCIL HEARING”x.com
- SUPPORT@sharongoldman“could attend a more balanced hearing”x.com