Stella Biderman

@BlancheMinerva

Ensuring that tech companies don't have a monopoly on being able to do research on cutting edge AI @AiEleuther. She/her

stellabiderman.ai

Joined May 2019

928 Following

19.3K Followers

Replies Reposts Media

Pinned
Stella Biderman
@BlancheMinerva
Aug 7
If you follow me and haven’t listened to this talk in its entirety, doing so is more important than whatever it is you are currently doing. There’s a lot of cybersec jargon at times but I don’t think you need to know anything about security to understand what’s important.
Eric Wallace
@Eric_Wallace_
Aug 6
Yesterday, my OpenAI collaborator and I gave a detailed talk on the Huggingface incident, our models creating "the message board", model misalignment, and more. youtube.com/watch?v=87DyyM… I hope it can answer a lot of the questions folks have, and we will release a full detailed
7
Stella Biderman
@BlancheMinerva
7h
We publish A at a workshop. We collab w/ people to write B, which goes on arXiv We update A and submit it to ICLR Reviewer calls us out on claiming novelty for things that were in B. They were also in the workshop version and are in B b/c of us How would you approach this?
15
Stella Biderman
@BlancheMinerva
Nov 4
If the Democrats do an above average job of being leaders and are willing to put being real liberals and making the world a better place as their #1 priority it would be the end of the Republican Party as a political force in the US.
2
Stella Biderman
@BlancheMinerva
Oct 31
It’s even worse than that… in the context of pentesting most users do not have permission to share the exploit prior to it being patched. This is a fundamental tenant of pentesting. You are automatically in violation of the terms of every bug bounty in the world all of a sudden.
Sean Heelan
@seanhn
Oct 30
haha, so apparently OpenAI are about to start training on your Daybreak Blue sessions, with no way to opt out. Literally stealing people's private work, as they do it, is cartoonishly evil. It also will presumably kill Daybreak usage by security researchers. Or at least, it
4
Stella Biderman
@BlancheMinerva
Oct 30
Actually this is why we carefully curate all LLM training text to be pre-2022, to avoid poisoning our models with dirty LLM outputs.
Riley Goodside
@goodside
Oct 29
We’ve known since 2023 AI could not continue to improve due to “model collapse”—training on poisoned internet data generated by AI. This is why LLMs have made no progress since that time, steadily declining on all benchmarks, and are projected to disappear completely by 2030.
4

Stella Biderman

Stella Biderman’s posts