4 ms·
Good point. This calls to mind LeCun's recent argument about missing models that can learn from raw experience or "self-play". When we have a ChatGPT that under
by intalentive 3y ago
Good point. This calls to mind LeCun's recent argument about missing models that can learn from raw experience or "self-play". When we have a ChatGPT that understands language strictly from audio / video inputs, then we can start to talk about human-like intelligence.
As for AlphaGo, I would put it the same category of intelligence as a calculator. It does one thing well -- approximate a Monte Carlo Tree Search.
- sharemywin 3y agoisn't HuggingGPT and Jarvis able todo some of that. https://github.com/microsoft/JARVIS https://github.com/microsoft/JARVIS