I'm very optimistic for near-term AGI (10 years or less). Even just a few years ago most in the field would have said that it's an "unknown unknown", we didn't have the theory or the models, there was no path forward and so it was impossible to predict. Now we have a fairly concrete idea of what a potential AGI might look like - an RL agent that uses a large transformer. The issue is that unlike supervised training y…
> Now we have a fairly concrete idea of what a potential AGI might look like - an RL agent that uses a large transformer. Any resources on that? I have a feeling that RL might play a big role in the first AGI, too, but why transformers in particular?
A Generalist Agent
Inspired by progress in large-scale language modeling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment generalist policy. The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens. In this report we describe the model and the data, and document the current capabilities of Gato.
Gato is a 1 to 2 billion parameters model due to latency considerations in real time physical robots usage. So for today standards of 500 billion parameters dense models Gato is tiny. Additionally Gato is trained on data produced by other RL agents. It did not do the exploration fully itself.
Demis Hassabis say that DeepMind is currently working on Gato v2.