Maxime Riché

Sep 29, 2024 | nuclear-news.net | Guillaume Amouret |Michalina Kowol |Maxime Riché |Christina MacPherson

Despite vastly different social and political contexts, Finland, Germany and France are all grappling with the question of safe nuclear waste disposal. “At first, there was strong opposition to the reactors, but it eventually disappeared”,……… One explanation lies in the massive financial support provided by the nuclear power plant operator, TVO, to the municipality of Eurajoki.

Aug 19, 2024 | lesswrong.com | Maxime Riché

Warning: This post was written at the start of 2024 as part of the AISC project "Evaluating Alignment Evaluations". We are not especially satisfied with the quality reached, but since we are not planning to work on it anymore, we are releasing it as a Work-In-Progress document. In this post, we cover:What are propensity evaluations? How do they differ from capability evaluations? Which propensities are currently evaluated?

Aug 6, 2024 | lesswrong.com | Maxime Riché

I am working on a project about estimating alien density, their expected utility, and arguing for strategic implications. I am optimistic about the project producing valuable content to inform the AI safety strategy. But I want to probe what the community thinks beforehand. Main question:Assumptions:Alien and Earth space-faring civilizations produce similar expected utilities . The utility is computed using our CEV.

Jun 10, 2024 | lesswrong.com | Maxime Riché

The words evaluations, experiments, characterizations, and observations are somewhat confused or confusingly used in discussions about model evaluations (e.g., ref, ref). Let’s define them more clearly: Observations provide information about an object (including systems). This information can be informative (allowing the observer to update its beliefs significantly), or not. Characterizations describe distinctive features of an object (including properties).

Apr 26, 2024 | lesswrong.com | Maxime Riché

My credence: 33% confidence in the claim that the growth in the number of GPUs used for training SOTA AI will slow down significantly directly after GPT-5. It is not higher because of (1) decentralized training is possible, and (2) GPT-5 may be able to increase hardware efficiency significantly, (3) GPT-5 may be smaller than assumed in this post, (4) race dynamics. TLDR: Because of a bottleneck in energy access to data centers and the need to build OOM larger data centers.