
Context: this is AI flagging of AI reviews without disclosure from reviewers. This is roughly a third of the reviews that I handle in my batch.

@CNRS Researcher

Context: this is AI flagging of AI reviews without disclosure from reviewers. This is roughly a third of the reviews that I handle in my batch.


We seek a math or cs PhD grad. Send us a joint email to apply. 📄 Full description here: files.samuelvaiter.com/posit…

📢 Postdoc opening! 📢 @n_keriven and I are hiring a postdoc to work on the theoretical analysis of large-context Transformers (infinite-token limits, depth, looped Transformers, reasoning traces). 📍Location: Rennes (IRISA) or Nice (LJAD), France

The alpha version of my new book "Optimal Transport for Machine Learners" is out, with in particular an online version with interactive figures gpeyre.com/ot4ml/

Of course you can always use complex step approximation if you don't like roundoff error and you have stable complex arithmetic :)
Complex step approximation is a numerical method to approximate the derivative from a single function evaluation using complex arithmetic. It is some kind of “poor man” automatic differentiation. nhigham.com/2020/10/06/what-…

A quick remark: When you use centered finite difference instead of forward or backward difference, the truncation error decreases faster, but I never realized that the roundoff error grows before!


So, yelling at a LLM to simplify a Lean proof really works!

Now accepted at ICML 2026! Congrats to @MagamedTm
🧵 New preprint! "Towards Understanding Steering Strength" with @MagamedTm and D. Garreau Activation steering is a popular way to control LLM behavior at inference. But how much should you steer? We provide the first theoretical analysis of the steering strength α. 1/7

TL;DR: Steering strength α has precise, predictable effects on LLM outputs. Understanding these laws may help design better steering strategies. Paper: arxiv.org/abs/2602.02712 Code: github.com/MagamedT/steering 7/7

We validate everything empirically on 11 models (GPT-2, Gemma 3, Qwen 3, Llama 2, Mistral 7B) across 8 safety-related concepts. All our theorems are confirmed experimentally. 6/7

Result 3: Steering always hurts global performance. Cross-entropy increases quadratically around α = 0, i.e., there is no free lunch. For large α, performance plateaus because the output becomes input-independent. 5/7

Result 2: Concept probability follows a sigmoidal curve. The probability that the target concept appears in the output increases smoothly with α, following a tanh shape. Off-target concepts decrease or vanish. 4/7

Result 1: Next-token probabilities have a bump shape. As α increases, most token probabilities rise, peak for α>0, then fall. In particular, off-target tokens peak before target tokens. 3/7

Take contrastive prompts (for ex, safe vs malicious), evaluate them through the model, compute the mean difference of representations at layer ℓ. That's your steering vector v. Then shift activations by αv at layer ℓ. Question: what does α actually do? 2/7