Here is Peter Wildeford on The Power Law and the implications of Claude Mythos, the most capable AI model yet. Its ability to identify and exploit cybersecurity vulnerabilities make it so dangerous Claude has decided not to release it. While this was the right decision, Wildeford argues that there’s currently a dangerous governance vacuum. As even more powerful models are developed, it’ll be crucial to address the security implications, and currently this isn’t happening. Another implication is that going forward, it may be that the most powerful models will not be publicly available.
Here is one of my previous posts on AI risk.
Nehaveigur 