|
| ▲ | tim333 6 hours ago | parent | next [-] |
| Re. disease cures I am hoping more for AlphaFold type stuff and simulating cells in silico rather than ChatGPT type LLMs. There is some progress like >“There are people sitting in our office in King’s Cross, London, working, and collaborating with AI to design drugs for cancer. “That’s happening right now.” https://www.htworld.co.uk/news/research-news/isomorphic-labs... and >...enables researchers to move seamlessly from AI-generated sequences to functional antibodies in just days https://the-decoder.com/googles-ai-drug-discovery-spinoff-is... |
| |
| ▲ | ted_dunning 18 minutes ago | parent [-] | | And none of that doesn't improve the throughput of the clinical trials. It just decrease the cost of coming up with things to put into trial. |
|
|
| ▲ | californical 6 hours ago | parent | prev | next [-] |
| But what if we could predict which treatments would be most successful with ~70% accuracy? It would potentially speed up the feedback loops right? There may also be downsides, like skipping testing things that would enhance our fundamental understanding of something because the AI was wrong. But that’s already a problem , and having a better gauge in the early stages could be really helpful |
| |
| ▲ | nradov 6 hours ago | parent [-] | | What if I could flap my arms and fly to the moon? You haven't presented any scientific evidence that LLMs will enable such prediction accuracy. It's pure speculation and hope. Some smaller, incremental improvements to optimize research workflows are much more likely. | | |
| ▲ | californical 5 hours ago | parent | next [-] | | I’m not saying that they will, but that investing in advancements to AI overall could do that. Not making predictions that they will, just trying to give an example of a benefit that we may get out of this | |
| ▲ | mckn1ght 4 hours ago | parent | prev [-] | | What is your opinion on AlphaFold? Doesn’t that provide a speedup for one part of medication development and understanding disease? | | |
| ▲ | yosame an hour ago | parent [-] | | Not really, you still have to validate the structures it estimates, which minimises any speed gains. It can help a little bit in the early stages of drug design, but even if it was perfect (which it's not), there's a massive gap between understanding a protein structure, and understanding how a drug will or system will interact with it. In a broader sense, understanding the structure of a protein is only a small part of drug development. Unfortunately biology is complicated, and we're an extremely far way away from solving it. |
|
|
|
|
| ▲ | Glemllksdf 5 hours ago | parent | prev [-] |
| LLMs help already a lot because plenty of normal people do not have programming skills. Evaluating test results is a lot harder if you do not know how to program or how to use a computer. But LLMs compute requirement is so high that it pushes the boundaries of compute, memory and memory bandwidth which is fundamental for curing diseases. LLMs math / neural networks can and are used for medical research. Simulating a whole body with proteins, cells etc. will bring us the breakthrough we need. Nothing in modern medicin research is withoout compute. AlphaFold def helps researchers around the globe. |
| |
| ▲ | nradov 5 hours ago | parent [-] | | More accurate biological simulations could help but there is zero reason to expect that LLMs will be an effective platform for such simulations. That's pure speculation and probably wrong. | | |
| ▲ | Glemllksdf 5 hours ago | parent [-] | | Im not betting on LLMs for this, i'm betting on the LLM Compute infrastructure which is the same for simulations. |
|
|
