This is the big issue, since a writable memory, a memory that can store arbitrarily long problems, and online learning procedures are too useful for the use case of advancing science significantly, especially cheaper.
Now, I do think we will be able to align even superhuman AI, but this is the issue I have with the proposition of narrowness: Science usually requires way more generality than most tasks.
This is the big issue, since a writable memory, a memory that can store arbitrarily long problems, and online learning procedures are too useful for the use case of advancing science significantly, especially cheaper.
Now, I do think we will be able to align even superhuman AI, but this is the issue I have with the proposition of narrowness: Science usually requires way more generality than most tasks.