Raw LLM Responses

Don't get me wrong, I think the only thing that will prevent humans from killing themselves is a Buck Rogers style AI government, but if I truly wanted to make an "Alphafold", I calculated it would take molecular dynamics simulations of refolding of all partial structures and unknown structures in solvent systems for at least 150,000 proteins as a training model, for it statistically to be able to really do what it says. In time terms, if I took Musks 10,000 GPU systems, the 10,000 that shorted a city, and ran 1 simulation on 1 GPU, if it has a 40% or greater structural homology to a known X-ray, or NMR structure for the entire amino acid sequence, it would take about 10 days. If there is none, ie no homology, it can take 6 months to a year. So 10,000 x say average 5% of proteins with a homology in that range would be around 5.5 months, x 15, just to collect the necessary training data. Then, each of these generate on the low end, after turning into spread sheets and images, 4 terabytes on the low end, and 100-200 terabytes on the high end...x 150,000 is the data storage capacity needed to hold the data for training, which is like 20 football fields cubed of pure 20 terabyte hard drives in arrays. My point. To me it just looks like google, and varied others made a deal with pharma companies whose 99% of money is small molecules, so as not to upset their 25% of all money on earth economics that drive the planet right now, and I have even seen the government of China make concessions to this for fear of a repeat of 1928-1945.