specked-citrus

I'm also quite surprised to see that LLMs can do this. I guess it is possible that "make an image in MS paint" is a type of RL environment used for image understanding. This is one of those areas where people inside the labs have a very different view into how much models are generalizing.

Add WYSIWYG site builders to the list of businesses on the chopping block.

Sooner or later, agents must be connected to the real world to do economically useful work. The sandbox failures are bad, sure, but the underlying misbehavior is the root problem here.

It's one thing to identify that these influence operations are happening, but another entirely to say they are a significant factor in public opinion.

[No system card yet](https://deploymentsafety.openai.com/)