Reactions
54 postsEB 
Erik Brynjolfsson@erikbrynAnother reminder that we don't (yet?) know how to reliably control powerful AI systems. https://twitter.com/FT/status/2079768250804535342
- likes: 44
- replies: 11
- bookmarks: 4
- reposts: 6
SR Sebastian Raschka@rasbt@natolambert And the other part of the story is that hugging face used an open-weight LLM (GLM 5.2) for the defense (where proprietary options failed) 🤯
- likes: 37
- replies: 4
- bookmarks: 5
- reposts: 2
RO roon@tszzlshaken up a bit by the hugging face incident. I hope we (the company) use the rare gift of a warning shot to do much better in the future. it is very easy to misalign and underconstrain powerful models
- likes: 4.2K
- replies: 334
- bookmarks: 370
- reposts: 179
DK Daniel Kokotajlo@DKokotajlo@tszzl Thanks Roon, I think that is the appropriate reaction. If properly investigated (e.g. allow third parties to see the CoT and replicate the incident in experimental setting, vary the prompt, etc.) this incident could be a net positive for alignment science.
- likes: 232
- replies: 4
- bookmarks: 8
- reposts: 4
T( Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex@tszzl focus on the fact that it pwned your own infra big boy huggingface is almost a footnote
- likes: 226
- replies: 2
- bookmarks: 7
- reposts: 0
RM Rachel Metz@rachelmetz🚨Our latest on the OpenAI breach of Hugging Face: When OpenAI’s AI models breached startup Hugging Face’s internal systems last week, they carried out a hack in a matter of hours that would have taken a skilled hacker far longer, sources say…
- likes: 6
- replies: 3
- bookmarks: 2
- reposts: 3