Source: Twitter X

Absolutely false. In fact the research and development is reverse engineering the brain and its intelligence by brute force working backward from language. They’re already working on wayfinding and adversarial selection (reasoning) the beginnings of which will be delivered this year; pseudo-parallelism this year or next; and neuromorphic architecture that will both produce that parallelism and end the cost and delay of pre-training. At least one group brute forced world modeling the physical world and embodiment by the same method. And my work on homeostasis and ethical and moral judgment and limits is computable. They have solved attention already, and consciousness, despite the laments of theologians and philosophers, is just a memory effect of all of the above.
The problem with the maturation of that combination of techniques, which I’m sure you’ve expressed in your statement of limits, is that the primary limit to demonstrated intelligence consists of (a) working memory – which of course, even the current tech can solve, (b) regulating ‘hallucination’ to creative ideation that’s then falsified by recursive wayfinding for operational possibility, and (c) the material and economic problem of hypothesizing (‘hallucination’), and constructing tests and experiments in the real world.
I suspect you have already noted that while there will be some early returns on the language models from synthesis of existing knowledge, and further from well established dynamic systems such as chemistry and proteins, but those returns are likely subject to diminishing returns – and we are left once again to imagining theories that can be tested by costly real world means. That is, until we have discovered sufficient first principles (laws) at every emergent layer of complexity (the scientific disciplines) that we’ve discovered the equivalent of the programming langauge of the universe.
Even then, knowing what is, is a smaller set of concepts than knowing what might be – which like language is infinite.
Cheers

Reply addressees: @GaryMarcus