Raw LLM Responses

As somebody who has worked in medicinal chemistry, I'm surprised the AI ONLY came up with 8.9 million designs. There are almost infinite possibilities for drug compounds (even considering that there are many criteria that they have to meet). This kind of thing gets tried pretty often and has a lot of problems that need to be sorted. Are the compounds the right shape? Size? Are there enough flat carbons? How many nitrogen/oxygen atoms are there? How many hydrogen bonding interactions are there? Are they soluble in water, but not TOO soluble in water? Are they toxic? Are they biologically active at all? Can they be cleared by the liver/kidneys without damaging them? Can they be synthesized efficiently (or at all)? The list of requirements goes on and on. It is pretty trivial to come up with an infinite list of possible structures, but it is actually a massive problem trying to figure out which compounds in that list have any chance of being any good. AI is potentially a great tool for solving this problem, but I don't think we're quite there yet. Good to see people working on it though.