Most of us right now are debating whether we have reached AGI, but I have a more interesting debate. I wonder whether we have reached ASI.
The thing is, the AI world is really scattered. To some extent, we have AGI, and to some extent, we have ASI.
Take coding, for example. I think we have reached AGI, but not ASI. Then look at mathematics and research. I think we have ASI in certain aspects. I mean, hundreds of problems are getting solved by AI, including problems that were previously beyond human capabilities. That is an example of ASI-like capabilities.
Then we have cybersecurity. AI can identify vulnerabilities and find ways into systems that were previously difficult to imagine. In certain aspects, we are seeing capabilities that look like ASI.
Then, for some stuff like image creation, we are still closer to narrow AI. That is not necessarily AGI or ASI, but a much more limited level of capability.
So, the AI world is really scattered. It's not that simple, where either AI is AGI or AI is ASI. We have different domains, and the AI capability pie is not evenly split.
It would be much more interesting to look at which domains AI is good at. Where do we have ASI? Where do we have AGI? And where are we still dealing with narrow AI?
I don't think everything will stay at a similar scale, and we cannot just define AI as one thing: AGI, ASI, or narrow AI.
Because it's really interesting when you look at it. We are really far ahead in certain areas, around human-level in others, and really far behind in some.