4 ms·

You remember the "cloud-to-butt" extension? I get the feeling that if we replace "AI" with "statistics" in similar fashion, we'll come much closer to the truth

You remember the "cloud-to-butt" extension? I get the feeling that if we replace "AI" with "statistics" in similar fashion, we'll come much closer to the truth of the matter in such articles though they sound far less exciting and sexy.

PaulHoule 3y ago
I’d say computational molecular biology has been over promising and underdelivering for a long time, when I was in grad school they were recruiting heavily for a program to determine how proteins fold in water that failed because proteins don’t fold, they get folded. I’m glad they finally took down the poster for the conference (held almost 30 years ago) just a few months ago in my building. In recent years these new approaches informed by ML are making a lot of progress finally. But there is so much fatigue on AI after just a year or two of revival one is tempted to not use the phrase at all. (Also I suspect that 90% or more of folks, even at a place like HN, is not aware there is anything more to AI than “Ask ChatGPT” but there is.)
- Wowfunhappy 3y ago
  > when I was in grad school they were recruiting heavily for a program to determine how proteins fold in water that failed because proteins don’t fold, they get folded. But didn't Google Deepmind solve the "protein folding problem" a few years ago, using AI?
  - odyssey7 3y ago
    It reads as if the distinction might be simulating a folding process in silico vs. predicting folds that would happen in vivo.
- fxtentacle 3y ago
  I've seen great results in computational molecular biology back in 2013. For the first time, there was enough computing power to run molecular dynamics simulations and/or to run linear regression on large molecule databases. The novel tests and medical treatments based on that exploration are mostly still in testing and certification, but I'd expect that in a few years, you'll see, for example, a new early blood marker test for sepsis, which my university patented based on Gromacs simulations roughly 10 years ago.

PaulHoule 3y ago

I’d say computational molecular biology has been over promising and underdelivering for a long time, when I was in grad school they were recruiting heavily for a program to determine how proteins fold in water that failed because proteins don’t fold, they get folded. I’m glad they finally took down the poster for the conference (held almost 30 years ago) just a few months ago in my building. In recent years these new approaches informed by ML are making a lot of progress finally. But there is so much fatigue on AI after just a year or two of revival one is tempted to not use the phrase at all. (Also I suspect that 90% or more of folks, even at a place like HN, is not aware there is anything more to AI than “Ask ChatGPT” but there is.)

Wowfunhappy 3y ago
> when I was in grad school they were recruiting heavily for a program to determine how proteins fold in water that failed because proteins don’t fold, they get folded. But didn't Google Deepmind solve the "protein folding problem" a few years ago, using AI?
- odyssey7 3y ago
  It reads as if the distinction might be simulating a folding process in silico vs. predicting folds that would happen in vivo.
fxtentacle 3y ago
I've seen great results in computational molecular biology back in 2013. For the first time, there was enough computing power to run molecular dynamics simulations and/or to run linear regression on large molecule databases. The novel tests and medical treatments based on that exploration are mostly still in testing and certification, but I'd expect that in a few years, you'll see, for example, a new early blood marker test for sepsis, which my university patented based on Gromacs simulations roughly 10 years ago.