Ensuring that tech companies don't have a monopoly on being able to do research on cutting edge AI @AiEleuther. She/her
Joined May 2019
- Tweets 13,672
- Following 863
- Followers 18,116
- Likes 12,145
696 Photos and videos
Pinned Tweet
In film, "we'll fix it in post" is what you say when something went wrong on set and you don't want to redo it. AI research has made it our entire methodology: train the model, then patch whatever comes out. Our new ICML oral argues this can't be the basis of a science of AI. š§µ
6
48
341
42,653
ICML Mech Interp had a lower acceptance rate than ICML this year.
10
2
158
18,366
Sorry for the confusion! I was referring to in-person papers, not all papers. I also fucked up the math in this thread, itās ~24% in-person Mech Interp and ~26% ICML.
1
5
908
Stella Biderman retweeted
My Fable 5 access is back!
135
247
7,650
646,393
Stella Biderman retweeted
For anyone looking for a quick primer on the 1st amendment as applied to model weights for oh I dunno no reason, @CenDemTech's comments in the NTIA's open weights proceeding from a couple years ago are a good place to start. See pp. 33-40. cdt.org/wp-content/uploads/2ā¦
Jun 13
WOW. This is an intense andāI would argueāunconstitutional escalation by the government, violating Anthropicās 1st amendment right to offer this information service and individualsā right to access it in the US. Time to litigate the right to access AI?
1
7
21
3,378
In film, "we'll fix it in post" is what you say when something went wrong on set and you don't want to redo it. AI research has made it our entire methodology: train the model, then patch whatever comes out. Our new ICML oral argues this can't be the basis of a science of AI. š§µ
6
48
341
42,653
Also, appolgies to @learning_mech and the "There Will Be a Scientific Theory of Deep Learning" team for not engaging with the contents of your paper. I believe I learned about it on the same day we got the ICML acceptance notifications.
1
1
9
1,992
From what I've skimmed I think we're in agreement about a lot of things, but I'm excited to find time to read it closely :)
5
1,496
Part of why post hoc analysis dominates: it's the only thing most researchers CAN do. Almost no one releases intermediate checkpoints or training data. we built MultiBERT and Pythia to set a better standard, and it's been great to see work like OLMo and Marin follow our lead.
1
3
19
1,212
Read the full paper: arxiv.org/abs/2606.06533 or come listen to our oral @icmlconf!
Huge thanks to my co-authors @aflah02101 @niloofar_mire @linguist_cat @FazlBarez @nsaphra
Stay tuned for a related workshop (hopefully) at NeurIPS too!
1
8
51
2,967
Stella Biderman retweeted
šØ As AI models improve, many benchmarks are becoming saturated and losing their ability to distinguish between models. šØ
Check out our new @icmlconf paper: āWhen AI Benchmarks Plateau: A Systematic Study of Benchmark Saturationā
3
15
52
16,926
Any reasonable analysis of the past five years makes it abundantly clear that major AI companies are making the world a worse place and open source AI isnāt. So they distract you from the actual harm theyāre doing in the world with hypothetical fearmongering about the future.
Jun 10
New blog.
I looked into the actual evidence and what models where used by bad actors to see whether closed models are safer.
Turns out: Nope, they are used to hack, misinform and scam. There is one exception, though.
Link in replies.
3
5
59
2,113
Incredible timing
Jun 10
CAISI has reportedly been directed to stop publishing public model assessments as the new AI EO gets implemented.
Natsec engagement on AI is essential. But pulling CAISI's evals from public view doesn't make the field more secure. It just means fewer eyes on the science when we need more.
Openness and natsec don't have to be in tension here. We should be doing both.
1
9
1,344
DoD: āAnthropic is a supply chain riskā
Anthropic: āYou aināt seen nothing yet.ā
Jun 9
mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community
also the fact that this is un purpose not visible to the user is crazy
10
26
371
49,135
Stella Biderman retweeted
Jun 9
They didnāt mean pause AI research, they meant pause *your* AI research
50
347
5,037
103,650
Itās super weird to me that thereās so much discourse about whether Anthropic is āconsistent.ā
Anthropic is choosing to make decisions that make the world a significantly worse and potentially more dangerous place. Thatās what you should criticize them for.
8
15
303
7,231