Form: Reply

Bad Bet Roko. Yes, it’s overhyped. But … the concepts in the A* algorithm applied to LLMS as Q* will produce and must produce competitive recursion instead of the one-shot algorithm used today. ( How long it takes them to tune it is another thing. )

The reason? The brain is vast but quite simple in practice: Memory > Episodes > Relations between episodes > wayfinding from episode to episode > recursive wayfinding to discover the best episode. All life that that has the equivalent of a nest, den, home, hunting or gathering range, and certainly all life that engages in any social coordination, evolved from this simple (trivial) process (algorithm) that requires nothing more than a hierarchy of memory (hierarchical recursive memory). And consciousness is the deterministic result of sufficient hierarchical recursive memory.

If successful, and I can’t see how it couldn’t be, then the pareto rule will apply – meaning that 20% of the development effort will include producing that set of faculties. The other 80%, like all humans, will be training it, so that it questions all the ‘sentences’ that have been used to program it’s networks.

Reply addressees: @RokoMijic