
post
Explaining AI to Americans: Hallucinate a burger.
AI trained on burgers plagiarises most prolific burger. Wow
Yeah, because the Big Mac is...so unique in its ingredients?

Using round balls as a model system, a generative AI system rediscovers the classic round ball without explicit supervision and generates novel round balls optimised for shape, colour, or diameter, according to a paper in npj Science of Balls and Testicles.
trying to explain optimization to a generative AI: imagine a burger
Yeah like wtf why is generative AI part of this. This is standard reward function optimization from the 1970s. Throw some simulated annealing at it and check in after 30 seconds
LLMs have limited uses but they're the only thing the entire western tech industry has decided to work with so we have to rebrand old, simple processes as being done by them
And even then, the only way we will know if these new, unheard-of burgers are actually delicious is to give them to focus groups to do taste testing, which is currently the process.
Do we really need an ai to tell us that "75 percent of respondents said they didnt like how sweet the bun was" means we shluld"look at reducing the sugar in the bun? I'm not sure what an AI is even doing in this process
Haven't you heard? LLMs can replace focus groups and polls because they're already made up of averaged opinions.
Thank god they haven't yet invented the RoboTongue
AI invents burger and all it took for it to do that is all the power of the Hoover Dam and the entire knowledge of human history including all the burgers.
"rediscovers"
Yeah, I rediscovered how to make a quarter pounder with cheese
Execs: We've finally found it, the AI of truth!
AI: Use fresh, high quality ingredients and lots of meat.
Execs: Nyehh!!
Yeah I ffucking love huge chunks of onion
(I don't and I actually think if you're putting slices of anything on a burger that are bigger than an eighth of an inch you're a lazy hack building slop piles)
I do like when my imaginary burger's imaginary feedback has a +30% Moist factor over the imaginary Big Mac's imaginary feedback.
Am I stupid or is providing a data set to model output definitely, unambiguously, 100% supervision?
"Delicious Burger 1" AKA the "Where's the Beef?" borger, going off of the AI slop graphic
What you've never eaten a burger and thought to yourself, "Man i wish this was just all pickles and onion and no meat at all"? Everyone who has ever been vegetarian knows the the goto meatless burger is just a shit ton of pickle and a 1cm thick solid chunk of onion. Talk about flavor!
From random samples, the model successfully rediscovers the classic Big Mac®, both in correct ingredients and weights, although the Big Mac® was never part of the initial training data (Fig. 2a). Across ten independent randomizations, rediscovering the Big Mac® requires on average 7.3 million samples which demonstrates that exact replication of recipes is a low-probability event under the learned distribution (Fig. 2e).
my siblings in christ that's not recreating the big mac, that's keep hitting your head against the wall untill you start to bleed lmao. i don't have anything specific to say against the paper (because i can't be bothered to look deeper) but this really specific and horrible to test benchmark is way too funny to not dunk on.
>create training sample built on millions of burgers that are derivatives of the big mac™, but use a different name for legal reasons
>the model produced a derivative of the big mac™
waow, these dipshits really indebted themselves to the state for their whole adult lives to do this, by the way.
but does it know about cutting boards and newspaper and high industrial gastopub burger experience? :tm:



Nature is healing
all 43 comments