Category: AI, Computation, and Technology

RT @SRCHicks: I’m an immigrant from Canada, and I’ve founded no engineering or technology companies. But this is still a cool chart and thr…

RT @Outsideness: Every day AGI gets closer, and every day I want it to kill more shit.

The monkey can’t. Because the correlation requires information external to the content.
LLM’s can’t because correlation requires information external to the content
Embodiment creates commensurability by determining senses (inputs) and cognition( disambiguation, world model (integration (perception)), episode(context, index), auto association(meaning), adversarial prediction, survival from competition, escalation(attention), and actions(outputs).

It can’t be done without Embodiment, world model, episode, and prediction from it. It might APPEAR so. But it would just be random generation.

That’s what’s required for creativity.
Folding proteins is just trial and error computation at scale.

However, given that LLMs are not AIs, but search mechanisms, that doesn’t mean we can’t produce AIs that convert speech into world models, predictions, permutations, wayfinding, and novel sequences of operations that ARE possible and ARE creative, which can then use language models to DESCRIBE those new possibilities.

The info isn’t in language.
The language produces episodes (contexts) which we can test against contexts.
There is a difference between linguistic consistency (what LLMs search for) and the identity, logical consistency, operational possibility, external correspondence, and cumulative coherence of all referents within necessary for ‘reason’.

Not sure how that applies to anything in the conversation so far, since LLMs are just probabilistic search systems of existing writing with randomness to seed branchings, so that they don’t produce repetitive results.