3 ms·
This may be true. But I think we should be open to the possibility that maybe not. What if, human intelligence, by and large is the product of developing our o
by DubiousPusher 4y ago
This may be true. But I think we should be open to the possibility that maybe not.
What if, human intelligence, by and large is the product of developing our own LLM. By this I mean, the logical circuits which are necessary for grammar and language structure are the basis for our generalizable symbol processing. We developed these neural network analogs to logical operations almost purely out of the benefit of being able to combine communication in a predictable way. We developed enough to achieve a kind of Turing equivalency. Essentially, we are a specific computer, that stumbled into an abstract machine.
Were that the case. It may be entirely possible to achieve general machine intelligence almost purely through an LLM. In which case, the LLM itself could potentially generate training data at a comparable quality to humans.
- theonemind 4y agoThe parts of our brain that work like an LLM do work like an LLM (hallucinates/generates falsehoods/over-confident/deals strictly with an abstract world of discrete symbols) and vastly overestimates its intelligence and how much it needs the rest of the brain's reality checking and other kinds of intelligence to turn it into something useful and conflates what it does with intelligence. It's basically a super-useful co-processor in the brain hardware, but it is a co-processor and not capable of independent operation.
- throwaway1851 4y agoYes. When I read comments such as “humans just predict the next word too,” I wonder if those commenters have ever stopped and observed their own thought processes before.
- DubiousPusher 4y ago> “humans just predict the next word too,” Just to be very clear, this is not what I said at all.
- throwaway1851 4y agoSorry. I wasn’t characterizing your comment; I was going off on a bit of a tangent.