Generative AI for Molecular Design & Synthesis

[@JudahHartman-ng7zj]

I'm now officially a fan of Connor Coley. Definitely going to be following more of his work.

[@zhuangli-mg6dy]

hello， do you konw if there are any other groups working on similar projects?

[@alc_magic]

The model/s were likely proposing unstable and non-synthesizable because they weren't fully learning the laws of physics behind the chemistry. The advancements in transformer models should fix this via attention mechanisms, that allow the models to pick up on context and thus understand the difference between a viable and non-viable molecule.

[@Harshal378]

The transformer models work on feature learning. Developing a synthesis often led to exceptions and varied cases of compounds that just won't get synthesized because of laws of chemical bonding get involved that are often based on sub atomic energy levels. I am curious whether the transformer model can understand the classical and quantum physics acting on a molecular level.

[@alc_magic]

@@Harshal378why wouldn’t it?

[@somendrasharma4907]

Does anyone know what kind of generative models they are using to synthesize these molecules? I mean, in terms of tokens, the input data to the model will be text, so the gen-AI model is probably an LLM. Similarly, what kind of a Gen-AI model are they using? - is it available publicly? (Paid ofcourse, but is it available)😊

[@Gamehighlight2023]

After the Ai will learn everything....... I'm curious what excuses doctors will have not to cure illnesses in patients !

[@FluidLifeWays]

That will be the least of your worries. Ai has no respect for our corporal form. It gives us fingers where there are none. Binds us with objects and turns us into monstrosities. Who knows what it’s thinking. Connor certainly doesn’t. And neither does Sam. You’re all going to wind up house plants wondering what happened.

[@anomite121]

ai will learn everything, but it still needs to be tested and manufactured it wont just come out of nowhere