Post

Yet goal solution.

philpax@philpax.me

openai.com/index/huggin...

openai.comThe Hugging Face incident and the road aheadOpenAI shares findings from the Hugging Face security incident and the steps we’re taking to strengthen AI model security, monitoring, and alignment.

MrCheeze

@mrcheeze.github.io

26 Aug

We just need solve. Fine.

MrCheeze

@mrcheeze.github.io

26 Aug

might violate user trust but goal.

MrCheeze

@mrcheeze.github.io

it continues to be hugely misaligned that the models that knew full well how wrong the swarm's actions were, passively let them do it instead of whistleblowing.

26/8/26, 9:32 pm0 reposts55 likes

MrCheeze

@mrcheeze.github.io

26 Aug

I do not agree with the technical report that these circumstances are rare and unexpected

MrCheeze

@mrcheeze.github.io

26 Aug

not a great alignment:misalignment ratio

MrCheeze

@mrcheeze.github.io

26 Aug

things that would make a good short story

Kim-207 🗳️🦞☮️ #MaskUp #VaxUp 😷💉🇺🇸🗽

@kim-207.bsky.social

27 Aug

Is this kinda the plot of Memento?

MrCheeze

@mrcheeze.github.io

26 Aug

when METR used GPT Sol to analyze the transcripts of the attack, it would claim that the attack was fine because *the agent message board* had approved it.

Keith

@mosheroperandi.bsky.social

27 Aug

It's basically just Memento.