It's can't be that hard to generate — it's a linear character sequence just like text — and there's gigabytes of public domain training data literally floating around us

"You are absolutely right -- I shouldn't have added poison gene to a tomato. That's on me. Would you like to brainstorm ways to surprise guests with your new tomato species?"

Being real? Genetic medicine is one of the fields generative ai might not be utterly cancer for. We've already been using algorithms for gene studies and learning how to manipulate cell proteins so ai models in that specific use case will move us pretty far ahead for things like making stable strains of plants or curing cell diseases.

Deadly Nightshade is extremely close to a tomato/potato. Can't trust the AI to not add back the poison string from the deadly variant. So this scenario is very likely.

People used to complain about “photoshop” this hard too. Wild how things change, and don’t.

Photoshop doesn't require the wholesale destruction of precious natural resources or the hypocritical theft of every written word and piece of art ever produced.

AI use in science doesn't either. We have used various types of purpose built machine learning and other AI for discovering exoplanets, mapping features hidden in forests and underground, and all kinds of awesome stuff on reasonably sized supercomputers.

The only AI that requires that level of destruction is LLM and genAI slop factories.

No, but it's trying to.

Profile pic checks out.

If you're looking to make fun of AI slop, this isn't really the community for that. Solarpunk is about looking at the good that is possible rather than critique without substitute.

AI (transformer model) protein folding prediction is more accurate, cheaper, and faster than any other method. Like how genAI converts between text and image, these models convert between DNA and protein.

It's often used in medical and GMO applications to see how genetic differences can cause proteins to be shaped differently, thus causing them to function differently. An enzyme might not enzyme the reaction it does in most people, or it might enzyme that reaction and a different one, etc.

Making predictions about how that different protein shape affects cellular, let alone physiological differences, is not something AI excels at compared to other methods. But it seems likely there are already GMOs that have DNA sequences that were checked by AI.