spoiler: i am working on a small AI driven game that doubles as a security research tool as my weekend side project :)

artificially constraining yourself to “AI but no AI made in the last 6 years” is stupid.

also, i would gently remind that we actually have essentially *no idea* at a paradigmatic or mechanistic level how brains _or_ LLMs do information processing, and confident statements about lack of isomorphisms are misleading

there isn’t really any reason to believe that existing neurological structure as evolved on earth is special in any particular way, let alone that it has a monopoly on _possible_ cognition substrate, so simply pointing out that AI systems are different is not a sufficient argument

do you think that a system has to match existing wetware in order to be considered “intelligent” or demonstrate cognitive ability? seems like an anthropocentric bias error to me

some ppl just simply want to punch

its a very straightforward and evil way of making an enormous pile of money, the Moloch Shortcut that opens up between “huh this is kind of useful” tier AI and research-grade frontier AI

(zitron’s take, that is)

just look at the success of tiktok and shorts. that level of mind control is achieved by only optimizing the recommender! imagine if we instead optimized every video and audio sample, and with much more powerful models

the take is so bafflingly wrong. i not only think that it is easy for AI companies to turn and exploit (using ads, direct optimization, and so on), i think it’s somewhat credible to consider just the Sloptimization problem a literal _existential risk_, independently of autonomy risk or other ai risk

yeah basically it's simclustering

or google translate (where the impetus for the transformer was initially invented)

or youtube

“gen AI” is largely just a product and marketing term. the technology is the same, enormous ANNs trained with supervised, unsupervised, RL, …, just pointed at different objectives

Product Name Driven Ontology

he’s a tourist who showed up out of the dirtbag left sphere to capitalize on latent resentment towards tech broadly, the AI thing is just a flavor that makes it seem current and important

america bad and that’s why we should give up civil rights and give them to the american government

the grifter part explains the _why_ (he is selling a narrative, and not interested in anything resembling truth or sense making)

yes he is an obvious grifter

but you don’t even need this to properly anchor his takes

just look at his predictive record and error rate + correction rate. i don’t think he has made a single correct prediction, and every piece is full of errors that he refuses to correct.

coddled westerners constantly want to dismantle the pillars and eat the seed corn and we have to constantly explain why that is a bad idea