#claude
Worried about Anthropic's new "Mythos" model?
You've probably seen it go by these past weeks: it supposedly found more than 10,000 vulnerabilities on its own, in software that runs just about everywhere. So I dug in, a deep dive heading into the weekend.
Mythos (and really the whole AI trend) rewards systems that are well maintained and punishes those that aren't.
The full deep dive is here: https://jorijn.com/en/blog/claude-mythos-ai-vulnerability-discovery-explained/
#cybersecurity #ai #infosec #anthropic #claude #security #mythos #jorijncom
Ran into several cases right now where Claude completely dropped the ball, and Codex fixed it in one pass.
Such a shame the ecosystem around Claude is much better than it is for Codex.
I've been experimenting with MCP servers in Claude, which essentially allow you to connect data sources like Garmin and Strava to AI.
This week, I noticed that my heart rate was higher than usual at the same pace, so I asked Claude to analyze the data. To my surprise, it found a cause and supported its findings with ample data.
I'm not thrilled about needing to run more Z2 to recover from fatigue, but technologically, I find this _really_ cool.