Is inference time compute fundamental to decision-making, or just an artifact of massive pre-training?
We prove it's fundamental. More compute (same parameters!) provably solves tasks that are otherwise impossible.
Excited to present a talk and poster tomorrow about this work!
Does classical computation theory help explain the success of inference time compute in RL? We study this question in our #ICML2026 oral
We prove that policies with higher inference compute solve and generalize to a larger set of tasks. Empirically, we show that such policies
It's been 2 days in korea (1 in Seoul & 1 in Gangneung) and I'm already awestruck by the kindness of Korean people.
1/2 A bystander gifted me their portable charger unprompted, just because they saw me struggling to charge my phone at a public charging booth.
2/2 Another
🇰🇷I will be at ICML next week to present our on inference time compute in RL.
Reach out if you are interested to chat about inference time compute, planning, world models and RL!
Does classical computation theory help explain the success of inference time compute in RL? We study this question in our #ICML2026 oral
We prove that policies with higher inference compute solve and generalize to a larger set of tasks. Empirically, we show that such policies
Atleast one paper in my reviewing batch for the last 3 conferences is AI generated slop.
It is time that measures such as blog.neurips.cc/2026/06/02/ai-… be adopted more widely.
