Victor Lecomte

Timeline

1. 2019

   Wilber's bounds for search trees

   In 2019, he and Omri Weinstein show that Wilber's Funnel bound dominates his Alternation bound on every access sequence for binary search trees, and give a tight Θ(lg lg n) separation for some sequences. The paper answers a conjecture of Wilber.

2. 2022

   Composition complexity of majority

   In 2022, he proves with Prasanna Ramakrishnan and Li-Yang Tan an optimal lower bound of Ω((n/k) log k) on the number of local functions needed to compute majority as a composition. The paper appears at the 37th Computational Complexity Conference (CCC 2022).

3. December 2023

   Incidental polysemanticity

   In December 2023, he coauthors "What Causes Polysemanticity?" with Kushal Thaman, Rylan Schaeffer and others. The paper shows that polysemantic neurons can arise incidentally, even when there are enough neurons to represent all features in the data.

4. September 2024

   Backdoor defense and learnability

   In September 2024, he coauthors "Backdoor defense, learnability and obfuscation" with Paul Christiano, Jacob Hilton and Mark Xu. The paper introduces a formal notion of defendability against backdoors, defined through a game between an attacker and a defender.

5. 2025

   A PhD at Stanford

   In 2025, he graduates with a PhD from the Stanford theory group, advised by Li-Yang Tan. His Stanford research covers computational complexity and other areas of theoretical computer science before a pivot to AI safety research; before Stanford, he is a master's student in Columbia's theory group and an undergraduate at UCLouvain in Belgium.

6. May 2026

   Estimating MLP outputs without sampling

   In May 2026, he coauthors with Wilson Wu, Michael Winer, George Robinson, Jacob Hilton and Paul Christiano a paper on estimating the expected output of a wide random MLP over Gaussian inputs without running samples through the network. For sufficiently wide networks, the estimator reaches a target error with substantially fewer FLOPs than Monte Carlo sampling.

7. September 2026

   Optimal shallow circuits for majority

   In September 2026, he and Prasanna Ramakrishnan post "Optimal Shallow Circuits for Majority". It presents a construction, found by GPT-6 Astra, of depth-d circuits of size 2^O(n^(1/(d-1))) for any symmetric function, which settles the asymptotic AC0 circuit complexity of majority and matches Håstad's lower bound.

8. As of October 2026

   Researcher at the Alignment Research Center

   He is a full-time researcher at the Alignment Research Center, which works on a systematic and theoretically grounded approach to mechanistic interpretability.