Monsieur Phi

Il faut qu'on parle de Claude Mythos

This video is not sponsored and was made somewhat spontaneously. The speaker explains that they should be editing the final, very long episode of "Philosopher View," which they want to release before a crowdfunding campaign ends. This means they don't have much time to delve into the world of Large Language Models (LLMs) and Artificial Intelligence (AI), despite many significant developments, particularly with the release of Claude Mythos by Anthropic. Claude Mythos is described as a general-purpose LLM, an advanced version of Claude Opus. The speaker finds the name "Mythos" to be an interesting choice. This model was used internally by Anthropic to identify vulnerabilities and security flaws in various software and IT systems. The results were unexpectedly successful, with Mythos discovering thousands of previously unknown flaws in operating systems, browsers, and products from major companies like Microsoft, Apple, and Linux. Due to its powerful cybersecurity attack capabilities, Anthropic has, for the first time, restricted access to this new model, deeming it too dangerous for public use.