That ends up going in the DeepMind philopsophy of making AI, rather than the OpenAI philosophy. DeepMind makes AI that can defeat the best human Go players, OpenAI makes LLMs that are much more versatile but struggle with finishing a chess match without making illegal moves, never mind winning.
The other question on my mind: Cases where we have known decision pathways is exactly what the expert systems of the 70s and 80s were about. And from my understanding there were some successes, but we mostly found that it doesn't scale. Both encoding optimal decisions and encoding the decision space of things that can happen are huge challenges. LLMs might be genuinely better here, because they can just make things up as they go along, but they are also difficult to control
Makes sense re Deepmind. I also guess it is easier to train an AI using Deepmind approach on the Starcraft 2, Go, or other more closed systems than in the rather open business world with all those rules that need to be interpreted by a human judge in case of problems.
Good point re expert systems. I need to look up what actually made them fail, or how they are not scaling.
The other question on my mind: Cases where we have known decision pathways is exactly what the expert systems of the 70s and 80s were about. And from my understanding there were some successes, but we mostly found that it doesn't scale. Both encoding optimal decisions and encoding the decision space of things that can happen are huge challenges. LLMs might be genuinely better here, because they can just make things up as they go along, but they are also difficult to control