it proves that the algorithm is embeddable in a bigger transformer of ~similar architecture.

symbol manipulation in transformers is fp arithmetic?

That's an implementation detail. The behaviour of trained transformer models remains similar even if you quantise them to 4-bit floats, or make every floating point operation noisy. This model only works if you use double-precision floating point.

and he has had several meetings change his opinion of the other party for no apparent reason (eg zelensky

extrapolation is futile

> Architecture wise and also just from a code quality perspective I have yet to encounter AI generated code that passes my quality bar.

You should consider trying to using AI in a programming language that scores high in the AutoCoderBenchmark.

it's a good sign? We are having these debates, and it's aired in public?

> pattern matches the "idea shape" of words in the "idea space

it does much more than this. first layer has an attention mechanism on all previous tokens and spits out an activation representing some sum of all relations between the tokens. then the next layer spits out an activation representing relations of relations, and the next layer and so forth. the llm is capable of deducing a hierarchy of structural information embedded in the text.

not clear to me how this isn't "understanding".

thats unlikely. but they are awfully lot like turing machines (k/v cache ~ turing tape) so their architecture is strongly predisposed to be able to find any algorithm, possibly including reasoning

> still applicable belief that slavery was really ended by industrialization making abolition economically...

not crazy especially as slavery was supplanted by debt, which is in a way fractional slavery (minus the chattel part ofc)

you could probably autoclave it with a standard dental/tattooing autoclave (~500 USD and requires a gas stove)