Is ai just lossy compression algorithm with information retrieval function?

Maybe that is how all neural networks work.

Reply to this note

Please Login to reply.

Discussion

Therefore to come up with any new solutions to existing problems, it requires new information from outside of the system or randomness.

Randomness could be iterative. Loop of

"does this work?"

As we live in a world, neural network agents have possibility to retrieve information outside of their system. Therefore internal retrieval or randomness can be replaced with outside of the system retrieval function, utilizing best guess keyword search.

Therefore well connected agents with excellent outside of the system information retrieval ability and fast outside of the system data transfer speed can produce excellent results, but still, whatever is the size of the nn, it cannot exceed the intelligence of collective system, which equals to the intelligence of all information available with reasonable transfer speed.

Conclusion: 7b models are almost as good as 700b models, If 7b models IS well connected to data sources (eg. internet).

Neither can produce super human intelligence, because the way nns work, is compression rather than inventing new solutions.