Valerio Pepe

valeriopepe.bsky.social

18 December 2025 18:52

I'm really excited about this work (two years in the making!). We look at how LLMs seek out and integrate information and find that even GPT-5-tier models are bad at this, meaning we can use Bayesian inference to uplift weak LMs and beat them... at 1% of the cost 👀

Naomi Saphra

nsaphra.bsky.social

10 July 2025 17:55

For tons more detail, including the training dynamics of vestigial heuristics and the influence of different hyperparameters on OOD generalization, check out our paper! Led by VR Li and @jennykaufmann.bsky.social helped by @wattenberg.bsky.social @dmelis.bsky.social and me! arxiv.org/abs/2507.06445

Naomi Saphra

nsaphra.bsky.social

10 July 2025 17:54

🚨 New preprint! 🚨 Everyone loves causal interp. It’s coherently defined! It makes testable predictions about mechanistic interventions! But what if we had a different objective: predicting model behavior not under mechanistic interventions, but on unseen input data?

Martin Wattenberg

wattenberg.bsky.social

20 February 2025 22:00

Today we're launching a multi-lab open collaboration, the ARBOR project, to accelerate AI interpretability research for reasoning models. Please join us! github.com/ARBORproject... (ARBOR = Analysis of Reasoning Behavior through Open Research)