Co-founder and Chief Scientist, @TransluceAI, prev @MIT
Joined October 2015
- Tweets 1,730
- Following 924
- Followers 3,129
- Likes 5,829
339 Photos and videos
Sarah Schwettmann retweeted
Feb 17
Why does GPT-5.1 Codex score 6.5% worse than GPT-5 Codex on Terminal-Bench, with the same scaffold? š§µ
GPT-5.1 times out at ~2x the rate of GPT-5. Excluding timeouts, GPT-5.1 wins by 7.2%. We analyzed 256M tokens of traces and found this in under an hour. Hereās how š
2
15
74
10,130
Sarah Schwettmann retweeted
Feb 9
We trained diffusion models on a billion LLM activations, and we want you to use them!
New preprint: Learning a Generative Meta-Model of LLM Activations
Joint work with @feng_jiahai, @trevordarrell, @AlecRad, @JacobSteinhardt.
More in thread š§µ
32
192
1,435
221,569
Sarah Schwettmann retweeted
21 Dec 2025
Overall, I'm excited to see more people signing on to the bitter lesson, scaling-focused approach to understanding AI. This was the core technical thesis that led me and Sarah to found Transluce, and I hope others will join us in these efforts.
x.com/TransluceAI/status/184ā¦
23 Oct 2024
Announcing Transluce, a nonprofit research lab building open source, scalable technology for understanding AI systems and steering them in the public interest.
Read a letter from the co-founders Jacob Steinhardt and Sarah Schwettmann:
transluce.org/introducing-trā¦
3
23
3,265
18 Dec 2025
All @TransluceAI work that I described in my NeurIPS mech interp workshop keynote is now out! āØ
Today we released Predictive Concept Decoders, led by @vvhuang_
Paper: arxiv.org/pdf/2512.15712
Blog: transluce.org/pcd
And here's @damichoi95's work on scalably extracting latent representations of users from model internals: transluce.org/user-modeling
7 Dec 2025
We can train models on maximizing how well they explain LLMs to humans š¤Æ@cogconfluence paraphrased. Mechanistic Interpretability Workshop #NeurIPS2025.
1
17
88
9,985
21 Dec 2025
And you can find the slides from my mech interp workshop talk here! š
docs.google.com/presentationā¦
3
1,533
Sarah Schwettmann retweeted
18 Dec 2025
Transluce is developing end-to-end interpretability approaches that directly train models to make predictions about AI behavior.
Today we introduce Predictive Concept Decoders (PCD), a new architecture that embodies this approach.
2
33
165
36,640
Sarah Schwettmann retweeted
17 Dec 2025
I'm really proud of what our team at @TransluceAI has accomplished in the last year! Take a moment to read our end-of-year post to learn what we're up to, and please reach out if you're interested in supporting us!
17 Dec 2025
Transluce is running our end-of-year fundraiser for 2025. This is our first public fundraiser since launching late last year.
1
7
65
8,768
Sarah Schwettmann retweeted
4 Dec 2025
Today we are announcing the creation of the AI Evaluator Forum: a consortium of leading AI research organizations focused on independent, third-party evaluations.
Founding AEF members: @TransluceAI @METR_Evals @RANDCorporation @halevals @SecureBio @collect_intel @Miles_Brundage
6
53
171
89,809
Sarah Schwettmann retweeted
28 Nov 2025
Have you ever had ChatGPT give you personalized results out of nowhere that surprised you? Here, the model jumped straight to making recommendations in SF, even though I only asked for Korean food!
1
18
48
6,944
Sarah Schwettmann retweeted
26 Nov 2025
Independent AI assessment is more important than ever.
At #NeurIPS2025, Transluce will help launch the AI Evaluator Forum, a new coalition of leading independent AI research organizations working in the public interest.
Come learn more on Thurs 12/4 š
luma.com/i6ekd5s2
4
13
68
13,159
25 Nov 2025
My favorite part of @damichoi95ās new paper (alongside 2 new datasets!) is the scaled up investigator pipeline that directly decodes open-ended user representations from model internals
end-to-end interp is increasingly promising and I'm excited for more work in this direction
25 Nov 2025
What do AI assistants think about you, and how does this shape their answers?
Because assistants are trained to optimize human feedback, how they model users drives issues like sycophancy, reward hacking, and bias. We provide data methods to extract & steer these user models.
6
24
4,420
25 Nov 2025
Come say hi at #NeurIPS2025!
@TransluceAI is hosting a lunch event on Thursday where we'll discuss our recent work on understanding AI systems and where we're headed next. Would love to see you there š
24 Nov 2025
Transluce is headed to #NeurIPS2025! āļø
Interested in understanding model behavior at scale?
Join us for lunch on Thursday 12/4 to learn more about our work and meet members of the team:
luma.com/8kjfb378
1
1
8
963
25 Nov 2025
We've been thinking a lot about:
*what are the right measurements to make, and subroutines to automate?
*how can we equip the ecosystem to not only make those measurements, but make sense of them? and build collective understanding of AI in a rapidly changing, complex landscape
1
2
275
25 Nov 2025
Excited to share some of our progress in these directions during our lunch talks! You can also find me speaking about:
*scalable oversight indep evaluation @ the FAR.AI alignment workshop 12/1-2
*end-to-end interp pipelines @ the mech interp workshop 12/7
4
252
Sarah Schwettmann retweeted
20 Nov 2025
Is your LM secretly an SAE?
Most circuit-finding interpretability methods use learned features rather than raw activations, based on the belief that neurons do not cleanly decompose computation. In our new work, we show MLP neurons actually do support sparse, faithful circuits!
7
76
368
119,307
Sarah Schwettmann retweeted
19 Nov 2025
Transluce is partnering with @SWEbench to make their agent trajectories publicly available on Docent!
You can now view transcripts via links on the SWE-bench leaderboard.
3
14
43
7,647
Sarah Schwettmann retweeted
8 Sep 2025
You have to care
21
108
640
136,470
Sarah Schwettmann retweeted
14 Nov 2025
Can LMs learn to faithfully describe their internal features and mechanisms?
In our new paper led by Research Fellow @belindazli, we find that they canāand that models explain themselves better than other models do.
5
57
272
67,752
Sarah Schwettmann retweeted
25 Sep 2025
Weāre open-sourcing Docent under an Apache 2.0 license. Check out our public codebase to self-host Docent, peek under the hood, or open issues & pull requests! The hosted version remains the easiest way to get started with one click and use Docent with zero maintenance overhead.
26 Aug 2025
Docent, our tool for analyzing complex AI behaviors, is now in public alpha!
It helps scalably answer questions about agent behavior, like āis my model reward hackingā or āwhere does it violate instructions.ā
Today, anyone can get started with just a few lines of code!
2
13
81
11,097
Sarah Schwettmann retweeted
12 Sep 2025
Agent benchmarks lose *most* of their resolution because we throw out the logs and only look at accuracy.
Iām very excited that HAL is incorporating @TransluceAIās Docent to analyze agent logs in depth.
Peterās thread is a simple example of the type of analysis this enables, but we have already found much more striking examples. Weāre validating these results now, and excited to share more soon.
12 Sep 2025
OpenAI claims hallucinations persist because evaluations reward guessing and that GPT-5 is better calibrated. Do results from HAL support this conclusion? On AssistantBench, a general web search benchmark, GPT-5 has higher precision and lower guess rates than o3!
3
12
69
15,749