Content pfp
Content
@
0 reply
0 recast
0 reaction

notdevin  pfp
notdevin
@notdevin.eth
Sent to me by a close friend in the specialty chemicals industry. Formally trained as a microbiologist and currently runs an R&D facility for creating novel food and supplement molecules. This guy has invented tech for pharma, energy, renewables, and weapons
9 replies
2 recasts
12 reactions

Claus Wilke pfp
Claus Wilke
@clauswilke
I always take these statements with a grain of salt. This paper used absolute state-of-the-art AI methods and was still several person-months of work. And in the end it's just a proof-of-concept. AI is not magic (yet). https://doi.org/10.1016/j.patter.2023.100865
2 replies
0 recast
1 reaction

notdevin  pfp
notdevin
@notdevin.eth
He isn’t saying no more work, just that he’s compressed the time it took to do work. I worked side by side with him for years, his singular output was already formidable
1 reply
0 recast
0 reaction

Claus Wilke pfp
Claus Wilke
@clauswilke
Ok, that's fair. In the paper I linked we summarized hundreds of patents with ChatGPT. Doing this manually would have taken forever, and likely we would simply not have done it, or on a much smaller scale.
1 reply
0 recast
0 reaction

notdevin  pfp
notdevin
@notdevin.eth
I’ve still never gotten it to do something novel, have you? I can get it to fill in the blanks between two knowns if the blanks are within the known data landscape, nothing beyond though
1 reply
0 recast
0 reaction

Claus Wilke pfp
Claus Wilke
@clauswilke
Yeah I think that's roughly correct. Definitely had that sense with my recent AI art piece (not sure if you've seen it). Changing the material just pulls in the typical style in which the material is used, not how it could be used in principle. https://clauswilke.com/art/project/human-shapes-of-matter
2 replies
0 recast
0 reaction