The cheerleaders are the LLM AI future true believers. I imagine they are the same people that were telling us about how NFTs will change the world last year.
Llms do stuff that has value. I can use Rfdiffusion with motif scaffolding to create a fusion protein with units from enzymes that have no crystal or cryoem structures with as high as a 5-10% success rate!!!!!! That's absolutely insane! I only need to print 20 genes to have a chance of getting a variant that works. Literal orders of magnitude improvement. And if you don't have a good multisequence alignment for getting a good folks from alphafold? Pure llm based esmfold can fill in the gaps. Evodiff is out here generating functional proteins with disordered regions. Also, to bring it back to openai, if I ask chatgpt to write some code, it writes some pretty decent code. If I give it a bunch of PDFs and ask it to give me a summary, it gives me a summary. I don't buy the agi end of the world hype so a shift that means that we get more focus on d eveloping useful tools that make my life easier that I'm totally willing to pay 20 a month for? Yeah I'm down. Get this cool tech and product into a form that's easy to use and useful and keep improving it!
-I'm literally drinking a coffee I bought with bit coin right now.
-I was able to send large sums of money to my grandma in another country while paying a fraction of the fees going through banks
-It's a stable store of value for people in volatile countries with unstable currency
-It's an investment available to the small timers, normally only the wealthy have these opportunities
-It lets me pay artists for their art directly and bypass the corrupt middlemen
this is a forum coding so i have no idea what any of that biology mumbo jumbo means, but everything you mentiond about chatgtp is conveniently missing a lot of details.
>write some code, it writes some pretty decent code. Is it trivial code? Is it code that shows up on the first page of any search engine with the same terms?
>it gives me a summary. Is it an accurate summary? Is it any better than just reading the first and last section of the report directly?
In terms of writing code, yeah it's pretty simple code. I'm paying 20 bucks a month not 200k a year. I've found it really useful to dive into open source code bases for papers (just upload the repo and associated paper) - academics write pretty garbage code and even worse documentation. It's able to easily and correcttly extend modules, explain weird uncommented and untyped code (what exactly is xyz data structure? Oh it's a tensor with shape blah where each dimension represents abc value. Great saved me 2 hours of work).
For the summaries - uhh yeah obviously the summaries are accurate and better than reading the first and last sections. Spend the 20 bucks and try it yourself or borrow someone else's account or something. Especially useful if you're dealing with nature papers and similar from journals that refuse to give proper respect for the methods section and shove all the information in a random way into supplementary info. Make a knowledgebase on both and ask it to connect the dots, saves plenty of time. I don't give a damn about the flowery abstract in the first part of the report and the tryhard conclusion in the last part of the report, I want the details.
It's comical that these useless hype bros can convince folks that a genuine computational breakthrough and a pretty solid 20 dollar a month consumer product with actual users must be bunk because the bros are shilling it, but luckily the baker lab doesn't seem to care. Can't wait to play around with all atom so I don't have to model a zinc atom with a guide potential and can just model the heteroatom directly in the functional motif instead! Not sure it'll work for the use case I have in mind until I try it out and print a gene or two of course but I'm glad folks are building these tools to make life easier and let me engineer proteins that were out of budget 3 years ago.
tell me... how much would it cost to develop a system that did this with pre-GPT OCR technologies? I know the answer. Do you?