The usefulness of alphafold had been grossly overstated by science media. The structures it generates are mostly nonsense, they typically can’t generate molecular dynamics simulation trajectories with the correct physical properties, which makes them not useful for drug discovery. in computational biology today you’ll most often find them in use as a starting point for structure refinement or as one of many sources of data for some integrative method. Again, not useless, and an improvement over previous structure prediction algorithms, but they didn’t ‘solve protein folding’ or anything, like Google might have you believe. I don’t put much stock in Nobel prizes, I mean, they gave one to Henry Kissinger.
Google also released alphafold 3 under a strict non commercial license, which severely limited its usefulness for both commercial and academic work, the latter of which all but requires using more permissive licenses. I assume Google wanted to monetize a commercial version of it at some point.
I’ve been using Openfold 3 in some work I’ve been doing recently, and while the software is a bit undercooked, it does produce structures similar to alphafold 3 without any of Google’s bullshit. So I wouldn’t worry too much about this announcement. The protein folding will continue with or without them
Though it does suck that attention is being diverted to LLM development because it’s so good at giving the illusion of being smart while still getting so much wrong.
Sounds like they just moved the research over to isomorphic labs, also even from the onset people quickly took the original approach and paper and made their own versions tuned for their needs.
Feels more like a research project having achieved its goals, the key players moved on, and they’re closing the Deepmind part of it as it’s not really useful anymore.
The next big thing in AI is “if LLMs reach sentience and AGI status, they’ll be doing even more great things”.
Google. The graveyard of projects.
So instead of continuing a project that could cure diseases and help understand human biology they’ll focus on a system I use for asking questions like “What adhesive is best to attaching siding trim?”
Well, you see, there are dozens of adhesive manufacturers willing to pay for advertising or top indexing, while any medical research that could incidentally mention gender or race has its funding set on fire by the US government.
“Finish nailer will get it through warranty and then it’s not our problem.”
- Home builders
It’s kind of a marvel that we all have the bulk of human knowledge at our fingertips, and instead of noble pursuits we largely use it for social manipulation and power grabs.
I guess there is just more population control in misinformation than there is in curing disease.
The radio was originally imagined to be a public education and cultural broadcast system. Instead it broadcast Amos and Andy.
always has been like that
That was the plan all along. Maybe people should (re) watch The Selfish Ledger
So the actually useful product gets shit down for the useless product
Useful for humans. Not for robots and reptiles.
amazing work
it’s crazy. some of this went from taking years to now taking minutes and people react negative because it’s using machine learning and AI.
Thank you Google.
Engadget changed the click bait title.
Google is not shutting down alphafold.
Jumper announced in June that he was leaving DeepMind to join Anthropic.
Key people were leaving.
Other former staff members were reassigned to Alphabet’s drug-discovery company Isomorphic Labs, which is a DeepMind spin-off.
It’s not as bad as the headline makes it sound.
Idiocracy
A tragedy has befallen humanity.
Correct me if I’m wrong, but didn’t AlphaFold basically solve protein folding, as in their master database of folds is now complete? Doesn’t it make sense to move on at this point?
We pursue AI for the good of humanity my ass
