Elektrine lite

← Feed

@wikiemol@mathstodon.xyz

2026-09-04 13:10 UTC

@Salty @tao to be clear I am not talking about “true super intelligence” or even anything that is “true AGI” (I don’t know if such a thing can exist or even is well defined in the colloquial sense) I am saying that training an LLM to _appear_ like an AGI or to make it _seem_ like it’s a “super intelligence” inevitably creates something analogous to a counterfeiting machine. It produces an unbounded amount of currency that is meant to represent real value, but the amount of actually valuable things has not increased. The amount of actually valuable things has not increased because of the data processing inequality. (Among other things, like the alignment problem) So there is a mismatch. I do understand Dr Taos point that where this analogy breaks down is that people do not have to use AI for “currency” production, even though it can be used for that purpose. But I think there are a lot of reasons why it is essentially guaranteed that people will at least in some domains, as long as information is treated like currency, which in our day and age, it is.

Replies (0)

No replies.