Real Science Is Harder Than Benchmarks: University students find that AI isn’t the cure-all in drug discovery, as AI isn’t as capable as human researchers for drug discovery and development

https://news.northeastern.edu/2026/08/26/ai-drug-discovery-study/

9 points · 6 comments · view on lemmy.world

6 Comments

Elting@piefed.social · 3 pts · 12d (3 replies)

Man, really turns out there isn’t a single field where the use of an LLM isn’t a huge compromise.

ViatorOmnium@piefed.social · 2 pts · 11d (2 replies)

Somewhat ironically, GenAI models are doing better than humans in mathematics, by the simple virtue of not getting bored or tired, combined with the oligarchs financing the AI companies being willing to spend infinite amounts of money until the agent shits out some new mathematical proof.

yeahiknow3@lemmy.dbzer0.com · 3 pts · 11d (1 reply)

This is actually ironic, because the one thing we know with absolute certainty that LRMs cannot do properly is mathematics. Turns out if you throw enough compute at boring problems with verifiable answers and a mathematician to check your work you get progress.

phailhaus@piefed.social · 2 pts · 11d

The verifiable answers part is important. There is a proofing language that allows complex math to be tested. LLMs are being used to brute-force that since they can be set up to run another iteration each time it fails.

SaveTheTuaHawk@lemmy.ca · 1 pts · 11d (1 reply)

It really doesn't help when people like Veritasium post misleading videos on how Alphafold "solves" protein structures.

themaninblack@lemmy.world · 1 pts · 11d

I heard that too from a doctor. I asked him if he’s worried about antibiotic-resistant pathogens.

“Not at all!” He said, quite enthusiastically. Because the problem had been solved.

He spent the rest of the night dodging specifics.

Strange fellow.