- OpenAI says it has put training of its largest model on hold while it upgrades monitoring, security and alignment, citing an incident involving Hugging Face and early signs that an upcoming model called Astra may reach a critical cybersecurity threshold. (Mitch)
- The announcement is read skeptically, with excerpts rewritten as Jurassic Park press notes and its wording tied to the Pacing the Frontier statement signed by more than 1,300 employees of frontier AI companies. (Mitch)
- Specific doubts include no mention of physically isolating untrusted models, a monitoring setup that can leave flagged activity running for up to an hour, and an alignment section that mostly promises more of what is already being done. (Mitch)
- OpenAI released ChatGPT for Teens, which watches conversations for signals that the user is a minor and moves them into a mode with opt-in parental controls, blocks on romantic talk, a study mode that withholds direct answers, and limits on claims of consciousness. (Mitch)
- The teen rollout opens a wider question about whether children need cognitive strain to develop, drawing on critical periods in language learning and asking which skills are worth cultivating in a world that may require fewer of them. (Mitch)
<- Back to digests
Daily digest
Life finds a way
By Mitchell Howe
OpenAI claims partial pause and rolls out ChatGPT for Teens
Read the full dispatch on AI StopWatch
From AI StopWatch, published with their permission.
The summaries and the full Spanish translation are produced automatically. The original is always linked.