First two things
There is fluid intelligence which covers ability to learn on the fly and adapt to brand new problems on the fly. Flexible and malleable.
Second there is crystallized intelligence which is the body of accumulated skills and knowledge that can rapidly solve or answer any particular challenge.
LLMs have very high amounts of crystallized intelligence. But it is still relatively brittle as seen in potemkin ai paper.
Now in the past many would have considered them agi. They can be viewed as a type of protoagi.
Given their superhuman knowledge an agi with such knowledge should be able to make tons of connections and discoveries by even cursory analysis of the body of knowledge. The information being stored in the proper data structure makes links evident.
So while current llms could be viewed as protoagis. An actual agi with superhuman knowledge, could be viewed as a weak super intelligence or protosuperintelligence.
A true superintelligence should be able to do things far beyond below human or even human level intelligence with superhuman knowledge. We are talking things like potentially intuitively mentally handling complex things such as protein folding or higher dimensional math in a way that no human can.
Some people have heard of synesthesia and how sometimes this aids in mathematics or other domains. In theory eventually a superintelligence may even understand the nature of qualia and develop a high level language based on qualia even qualia beyond human comprehension.
Like we humans craft uis and interfaces usually with colors and graphs to handle certain things an asi may craft alien qualia uis akin to synesthesia but with foreign unknown qualia beyond human senses. Such may aid in processing of domains beyond human comprehension. This hypothetical high dimensional language, and ui design space, Id call qualiange.
No comments:
Post a Comment