Alessandro Stolfo

1/6: Can we use an LLM’s hidden activations to predict and prevent wrong predictions? When it comes to arithmetic, yes!
I’m presenting new work w/
@alestolfo.bsky.social
“Probing for Arithmetic Errors in LMs” @ #ICML2025 Act Interp WS
🧵 below

18.07.2025 17:22 — 👍 1    🔁 1    💬 5    📌 0

Logo for MIB: A Mechanistic Interpretability Benchmark

Lots of progress in mech interp (MI) lately! But how can we measure when new mech interp methods yield real improvements over prior work?

We propose 😎 𝗠𝗜𝗕: a 𝗠echanistic 𝗜nterpretability 𝗕enchmark!

23.04.2025 18:15 — 👍 51    🔁 15    💬 1    📌 6

@vidhishab.bsky.social Safoora Yousefi @erichorvitz.bsky.social @besmiranushi.bsky.social

15.04.2025 16:36 — 👍 0    🔁 0    💬 0    📌 0

Our paper "Improving Instruction-Following in Language Models through Activation Steering” has been accepted to #ICLR2025!

We're also excited to share that our public GitHub repo is now live.
Code: github.com/microsoft/ll...
Camera-ready: arxiv.org/abs/2410.12877

15.04.2025 16:35 — 👍 8    🔁 2    💬 1    📌 2