Quote:
Originally Posted by BlackJackDegen
btw, I think reasoning of some sort will mark a huge pivot point for AI in general that will effect almost every area in society. I think of current AI trajectory, like the part when you are getting pulled up a roll coaster on chains and once you incorporate a decent level of reasoning is when you tip over the edge onto the actual roller coaster. The inference costs are the current main impediment from what I have read which the article above says Open AI have tried to tackle in various ways. Generally though we only think about inference costs in terms of having a model that works at mass scale. So what will be available to us as consumers.
The real question though is who cares about the cost if the model can solve real world problems that humans currently can't do that would completely transform the world outside of AI. So what that means is lets imagine the inference cost is huge. Who cares about the cost if it is giving cures for cancer, math problems, etc.
I didn't even know when I was making these posts we were so close to a model release with some level of reasoning. Open AI have released new models in the last few days with the introductions of some reasoning capability. They are smashing current benchmarks. Talk we will get even crazier models released before the end of the year.
I listen to most people with extreme scepticism when they start talking about crazy timelines around AGI. When its someone as sensible as Noam who is working at the cutting edge of the field you tend to pay more attention. ARC benchmarks are generally considered a good test of human level performance.