3 ms·

My first reaction was, "But what if the AI generates incorrect code, and the bio people don't realize it?" But then, without the AI, what if the bio people gen

My first reaction was, "But what if the AI generates incorrect code, and the bio people don't realize it?" But then, without the AI, what if the bio people generate incorrect code and don't realize it? Can the AI be more accurate than a non-programmer? Probably; in fact arguably they already are. So I think I buy the parent's point. In fact, I think it's wider than biology. I bet there are a bunch of other fields with non-programmers doing a fair amount of coding. Economics research, say. Business process automation. I'm sure there are others.

Balgair 1y ago
Yeah, just the stats got so bad in the less hard sciences that they released statcheck to help automatically combat it: https://pmc.ncbi.nlm.nih.gov/articles/PMC7540394/ https://pmc.ncbi.nlm.nih.gov/articles/PMC7540394/ Byand large, since bio people just do not like programming, they were commonly generating outright bad code and then spending just months/years trying to figure it all out, hopefully before publication. But not always. LLMs have democratized passable code. I think it'll be before 2030 that they'll have democratized good code.
wrp 1y ago
> ...what if the AI generates incorrect code, and the bio people don't realize it? Cynically, there are plenty of people who don't care so long as they can get it published. Pragmatically, the quality of work is frequently so poor, leading to so much random variation in results, I doubt that frequent errors in AI-produced work would significantly change the quality range in many fields.
_DeadFred_ 1y ago
If you think being in IT and supporting random Excel spreadsheets you never knew about and random shadow IT projects that you suddenly get ownership of is a pain. I can't imagine all the vibe coded projects about to come IT's way. With zero standardization on frameworks/DBs/languages used, zero written specs/requirements/anything. It should be fun.

Balgair 1y ago

Yeah, just the stats got so bad in the less hard sciences that they released statcheck to help automatically combat it: https://pmc.ncbi.nlm.nih.gov/articles/PMC7540394/ https://pmc.ncbi.nlm.nih.gov/articles/PMC7540394/ Byand large, since bio people just do not like programming, they were commonly generating outright bad code and then spending just months/years trying to figure it all out, hopefully before publication. But not always. LLMs have democratized passable code. I think it'll be before 2030 that they'll have democratized good code.

wrp 1y ago

> ...what if the AI generates incorrect code, and the bio people don't realize it? Cynically, there are plenty of people who don't care so long as they can get it published. Pragmatically, the quality of work is frequently so poor, leading to so much random variation in results, I doubt that frequent errors in AI-produced work would significantly change the quality range in many fields.

_DeadFred_ 1y ago

If you think being in IT and supporting random Excel spreadsheets you never knew about and random shadow IT projects that you suddenly get ownership of is a pain. I can't imagine all the vibe coded projects about to come IT's way. With zero standardization on frameworks/DBs/languages used, zero written specs/requirements/anything. It should be fun.