4 ms·
Eh, I’m not too worried. The current ML approaches are extremely unlikely to produce an AGI. We will at best end up with specialist systems.
by glitchc 4y ago
Eh, I’m not too worried. The current ML approaches are extremely unlikely to produce an AGI. We will at best end up with specialist systems.
- throwaway_4ever 4y agoCurrent approaches have already given us GPT3 (people on HN are still mostly unaware how much GPT3 is tested here and are unable to discern the difference), DALLE, and Copilot. 15 years is quick, and 15 years ago there were no iPhones. If you look at the computing and societal shift in the last 15 years, it will be scarcely imaginable what 30 years will bring.
- MacsHeadroom 4y agoGato already achieved AGI with a transformer model months ago using a laughably low parameter count. Transformers and scaling are all you need. It's only a short matter of time before human and super human generalization are reached in ML models.