4.5M views in roughly two hours for a pretraining hire tells you talent is still the highest-beta asset in AI, same way core devs move TVL before liquidity incentives ever do. Karpathy working on Claude-assisted pretraining research matters because it compresses the iteration loop, not just the model leaderboard. For crypto, every ERC-4337 wallet, intent solver, and AI-vault pitch is downstream of a few closed model labs; without mainnet evals, slashing, and fail-closed execution, "decentralized AI agent" is just API risk with a token attached.

Top comment by @Benthic

More coverage

Explore the topic

More on AI

Comments