I think curiosity is very likely to be an important factor in achieving AGI. Without a curiosity mechanism, it means the agent may never proactively engage in learning. Almost all animals in the natural world have a curiosity mechanism, and I believe this may explain some issues.
As impressive as they are, I do not believe LLM are the path to AGI. It might look like it, but I don't think it can... I'm a firm believer that RL is the way !
Reinforcement Learning is simply too fragile, none robust route, requiring excessive reward shaping. There is no clear way towards AGI, and we are no nearer to it, than when we believed we were within decade or so in the mid 1960s (Minsky et al). Avoid the AI hype.
I think curiosity is very likely to be an important factor in achieving AGI. Without a curiosity mechanism, it means the agent may never proactively engage in learning. Almost all animals in the natural world have a curiosity mechanism, and I believe this may explain some issues.
Great overview Phil! And good refresh on that Sutton paper (had forgotten about that one). Chris
Thank you so much Phil! Excellent overview.
As impressive as they are, I do not believe LLM are the path to AGI. It might look like it, but I don't think it can... I'm a firm believer that RL is the way !
Thank you :)
Reinforcement Learning is simply too fragile, none robust route, requiring excessive reward shaping. There is no clear way towards AGI, and we are no nearer to it, than when we believed we were within decade or so in the mid 1960s (Minsky et al). Avoid the AI hype.