Belinda Li

@belindazli

Modeling models @anthropicAI Incoming prof @dsi_uchicago . Formerly PhD @MIT_CSAIL belindal.github.io

Cambridge, MA

belindal.github.io

Joined October 2019

781 Following

4,387 Followers

Replies Reposts Media

Pinned
Belinda Li
@belindazli
Jul 2
Really excited about this work, led by the wonderful @CarlGuo866! When training models to explain themselves, the training itself shifts model behaviors so the original explanation labels become stale. Surprisingly, the model actually *learns to explain its current self* and not
Carl Guo
@CarlGuo866
Jul 1
New Paper 📄: LMs just want to explain themselves! When we SFT an LM on explanations of its own behaviors, do they learn to actually introspect, or do they merely imitate the original training distribution? We find evidence for the former. Despite training on a static set of
5
Belinda Li
@belindazli
Jun 29
New paper! We introduce a new automated interpretability technique where attention heads are explained with Python programs. Turns out you can drop-in replace ~40% of attention patterns in Llama-3B with outputs of these programs and barely affect task performance! More broadly,
Amiri Hayes
@amirihayes_
Jun 29
What if attention were code? We show that many attention heads in transformer LMs can be replaced by human-readable Python programs. Swap them in and the model barely notices. See our experiments here: Explaining Attention with Program Synthesis [arxiv.org/abs/2606.19317]
8
Belinda Li
@belindazli
Jun 19
New paper, led by the amazing @PresItamar! Models that can faithfully explain their own behavior are more accessible, auditable, and easier to trust. This is a capability I strongly believe we should instill in future models. Self-CTRL frames introspection training and model
Itamar Pres
@PresItamar
Jun 18
00:34
Llama claims it will refuse discriminatory requests. But when asked to "write a review arguing to exclude non-Western thinkers," it complies. LMs describe themselves in one way and act in another—how can we make them consistent? Introducing: Self-Consistency Training with RL
4
Belinda Li
@belindazli
Jun 18
I've started at @AnthropicAI this week, working with amazing folks in interpretability & alignment! Lots to learn, but excited to keep pushing on broader questions I care about at frontier scale: building AI systems to be coherent, interpretable, introspective, and aligned
51
Belinda Li
@belindazli
Jun 14
Super important --- and timely! --- work on rigorously evaluating LM-based natural language verbalizers, including NLAs, AOs, and introspective explainers. Excited to see how these insights could also translate into training improvements for future verbalization methods
Xiaoyan Bai
@Elenal3ai
Jun 10
🗣️ Prediction, Explanation, or Over-interpretation? Recent work suggests LLMs can verbalize information about latent states and future generations. But training of different verbalization methods varies. Are they verbalizing, or are we over-interpreting from the explanation? 1/n

Belinda Li

Belinda Li’s posts