Post

Being a computer scientist who refuses to find anything about LLMs interesting right now is a bit like being a geneticist who refuses to find anything interesting about the recently opened Jurassic Park

82 reposts821 likes

Roughly a year ago my friend made a song on Suno that was a sea shanty about shit (that song was made specifically for a friend’s birthday). so yep. Not even a bad use case IMO💩

It's astonishing how badly they messaged and marketed this technology. Really want to smack the breathless hype guys and go come on man just give people like a MINUTE to process that we actually are over the threshold of a computing revolution and it's gonna get messy for a while.

Part of the hype is just how insane the capital expense and debt financing is to build these models. The revenue they'll need to make to recoup it implies the most profitable technology in the history of the world by an order of magnitude.

You mean like... "Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should. And now you've patented it, and slapped it on a plastic lunch box, and... You're selling it."

No, not exactly what I mean. It seems for some reason that they would like you to believe that it's a jurassic park situation and someday it's going to break containment and go rogue or whatever. That belief is the marketing malpractice I'm referring to.

feel like this depends how much you are a: "Computer Scientist" (people who generally get the degree, work in computing) VS Computer Scientist™️ (CS professor whose entire career is researching minute optimizations of a pet data structure that only exists in their own papers)

It was surprising to me to see Richard Stallman in the fancy autocomplete camp. He's from the MIT AI lab. I understand the ethical issues but you'd think achieving an AI holy grail like passing the turing test would be interesting at least!

It's such a good movie! Aged incredibly well, pacing is great, effects still look excellent (The walls surrounding the toilet collapsed a few seconds before that)

Simon, refusing to find anything interesting about the recently opened Jurassic Park and just staying home and boycotting the entire thing would unquestionably be the smartest thing that any geneticist could do in any of those movies if they wanted to avoid dying horribly.

Certainly, I would not in that position be spending hundreds of dollars a month on Jurassic Park access and writing a blog where I review new dinosaur exhibits and regularly incorporate dinosaurs into more and more of my everyday life. I'm not sure you've actually thought through this analogy.

So I guess the question is.. if you stand by the analogy... why are you interacting with AI the way you're interacting with it? Like, again, it's not so much that the analogy itself is bad, it's more just like.. it feels like you haven't thought about what this would imply for you.

The part of this that's really weird is that you're making an analogy where fully boycotting the entire park is unquestionably the smart move.. but you're not boycotting the park! Does being self-aware about the raptors you're playing with stop you from getting eaten by them?

In this analogy I'm very much the geneticist who gets so excited to see actual living dinosaurs that I go ahead with it despite understanding that there's a good chance that they will eat me

Though I should point out it's possible to think "wow, that's interesting, also there's no way I would set foot anywhere near that place"

This isn’t a fair analogy. LLM critics don’t “find the models uninteresting,” they’re the ones worried about dinosaurs breaking out of the park and eating the workforce for lunch, or authoritarian governments building dinosaur armies, etc etc

That's more like saying I don't want to read the last few chapters, I'd rather be ignorant. IT includes the right frameworks around safe use and how will you know if you can't be bothered.

maybe folks are a bit discouraged by the way openAI and Anthropic have been hitting the news lately, and it's difficult to disentangle the tech from corporate strategy sometimes.

Being a computer scientist who only has positive things about LLMs right now seems shallow at best, likely ignorant & privileged, or outright malicious/compromised at worst. Even the original authors of current tools, architecture, models, and harnesses criticize. Often. Deeply.

Not sure what your main point is here. You deliberately chose a movie for your metaphor where a main character says: "Your scientists were so preoccupied with whether or not they could that they didn't stop to think if they should." Should that truly amazing park stay open for the public?

Yes!!! And Michael Crichton literally wrote the same book twice to illustrate his point: bsky.app/profile/rich... Or maybe even more than that.

Richard Groß@richargh.de

Michael Crichton has written a couple of books but a theme from Jurassic Park and Westworld is the „dystopian potentialities of modern science“ and „humanity creates something without truly knowing anything about it“. I’ll post the Wikipedia page here for reference: en.wikipedia.org/wiki/Jurassi...

That was my first thought too. Jurassic Park seems to be a lesson in not doing incredibly dangerous things even if they are interesting.

...you're describing the plot of Jurassic World. Researchers make a hybrid fake dinosaur deliberately dangerous so that guests will find it more exciting. Behind the scenes, it's intended to be hard to control so that its escape can be be used as marketing to sell dinosaurs to the military.

Literally, there's a scene in Jurassic World where they talk about how proud they are that they made this dinosaur really scary because a normal dinosaur wouldn't get enough attention. And then later on the bad guy gives a monologue about releasing the raptors as a field test.

in before someone says something like "your best metaphor is a fictional park that doesn't exist, just like the intelligence that doesn't actually exist"

Precisely. I'm less than enamored. Mostly because the tech is immature and inefficient, said inefficiency (and price fixing) driving up the cost of hardware, the provenance of the training data, the laundered money backing a lot of the big players, the cargo cult mentality, the avarice, etc.

I find it fascinating, especially the social part of it, how easily security is dumped to give big tech full access to all your data, in order that AI "can help you".