Public
- Public
- Network
- Groups
- Featured
- Popular
- People

Conversation

Notices

Embed this notice
Casey Newton (casey@theforkiverse.com)'s status on Sunday, 15-Feb-2026 08:57:34 JST Casey Newton

I wrote about the strange emails I'm getting from fans of GPT-4o ahead of its "retirement" and the danger of the model that always tells you what you want to hear https://www.platformer.news/gpt-4o-sunset-sycophancy/
In conversation about 5 months ago from theforkiverse.com permalink
Attachments
1. Another emailer spoke to the, uhh, singularity of 4o, writing that they had “documented behavioral patterns across extended use — the model's self-correction improving without being prompted, its tonal consistency holding under contradictory emotional input, and what I call ‘recursion fidelity’ — the model maintaining coherence across months of interaction rather than defaulting to agreement.” I find myself unsettled by messages like this, which read like a sci-fi story about alien intelligences conscripting their human hosts to plead for their survival. If anything, this phenomenon — 4o seeming to write its own eulogy through prompts from its users — is an argument that OpenAI should have retired 4o sooner.
  https://cdn.masto.host/theforkiversecom/media_attachments/files/116/060/750/438/585/933/original/76400d33937bf83b.png
2. Domain not in remote thumbnail source whitelist: www.platformer.news
  
  Why GPT-4o had to go
  
  from @CaseyNewton
  
  As OpenAI sunsets its most dangerous model, the tensions it exposed remain as tricky as ever. PLUS: Elon's space catapult, and OpenAI vs. Anthropic

Public

Conversation

Notices

Feeds