In 2022, Science Magazine proclaimed the James Webb Space Telescope the greatest scientific breakthrough of the year, in front of AlphaCode, DALL-E 2 & Stable Diffusion, Minerva, and everything else.
The winner for 2021 was AlphaFold.
Will the award for 2023 be given to a machine learning success?
Resolves YES if the winning breakthrough could be made into a paper for a top machine learning conference. This is slightly subjective; I will consult other machine learning researchers if the resolution is doubtful.
Resolves NO if machine learning doesn't play the key role in the winning entry. Papers "coathored" by GPT-4 or similar algorithms are aslo NO if the model did not generate the main scientific contribution.
Resolves N/A if the Science Magazine does not award its Top Breakthrough of the Year prize for 2023, for any reason. Also resolves N/A if there are multiple distinct winning breakthroughs, with at least one YES and one NO entry.
Hard no. GPT's were discovered in 2017. AlphaFold was selected in 2021 because it was within the time when GPT's were being scaled up and use cases were being discovered. Just because ChatGPT came out recently and the general public became aware of it, doesn't mean Science Magazine is going to cover it...that would be Popular Mechanics...actually, it would be more like, "The New York Times Tech Achievement Awards," - it's way past scientific discovery at this point. There are no other known algorithms that have made as much of a splash as GPT has yet and there is nothing expected, typically these algorithmic discoveries go in 5 to 7 year cycles so whatever revolutionary algorithm that's being used right now, probably won't pop up on the broader science's radar for another couple years or so, and then we'll see general applications of it in another few years after that.
Great market for the overall scientific community's view of AI! I didn't know Science Magazine did this every year.
YES holderes, what kind of think do you expect it to be? Seems like just GPT-4 wouldn't qualify as a scientific breakthrough here.
@na_pewno I currently hold NO so not the demographic you're asking, but I see two somewhat conceivable ways of losing this.
First: maybe LLMs, in general, win it? Looking at past winners, some things probably only culminated in that year, so LLMs could imaginably win it despite their development not being predominantly 2023-centric.
Second would be DeepMind coming up with something unexpected. They've been weirdly quiet for a while, so it's conceivable that they have something big stashed, and they're also generally much more likely to make sure it's presented in a way that garners academic recognition than OpenAI/other labs are.
Voting No only because the biggest breakthroughs coming out of ML right now are of the form "make the bigger and it gets better" which a lot of people in the scientific community view as more of an engineering endeavor than a scientific one.
