The real-time voice interruption, the laughter, the emotional tone shifts — it crosses the uncanny valley in a way that genuinely unsettled me. I was testing it for an hour and kept forgetting it wasn't a person. The latency is basically zero which is what makes it feel so different from previous voice AI. We are not ready as a society for this to become widespread.

Llama 3 70B is running on my local machine and outperforming GPT-3.5 Turbo on coding tasks by a significant margin. The fact that this is free, private, and runs offline is genuinely game-changing. Anthropic and OpenAI should be worried — the open source community is closing the gap faster than anyone predicted six months ago.