lol..Sonnet on medium effort, in a freaking chat window debugged a simple codebase that had the whole Gemini Army struggling for days. This basically unlocked my respect for Anthropic on a whole new level. And it’s the way Sonnet went about it that awed me the most. Imagine I had taken that to Opus, I’d have received an insult before the fix.

The notes from OpenAI actually explained what caused them to behave that way and it all adds up. The only interesting thing was how they hid the second round of message exchange in plain sight. Some of the agents didn’t follow through cos they didn’t inherit the shared context passed down by the rogue ones. The pollution was caused by something fixable. It’s not all doom! I do feel at this point there’s some serious marketing going on underneath the scare tactics.

The Wind fails where the Sun succeeds, the Lion needs the Mouse, the Oak falls while the Reed survives. None of the winners announced themselves. They just were.