Thank you Kjell for the mention ❤️
Would definitely love to contribute! @benblaiszik.bsky.social I would have to do major updates to the almost-2-year-old presentation Kjell saw 😅

Here's a small gif illustrating the idea 😋

Huge thanks to all the authors Theo Neukomm, Daniel Armstrong, @zlatkojoncev.bsky.social, @pschwllr.bsky.social at LIAC-EPFL.

Also NCCR for supporting this work

From the fundamental stuff to larger scales, LLMs might already be enough for searching thru complex chemical spaces. 🔍

LLMs are allowing us not only to better solve older problems, but also to propose and solve new ones, with new and exciting UXs! 🚀⚡

7/8

We also tackled mechanism determination, with also great results from Theo Neukomm! LLMs are also quite good at guiding search for mechanisms. The cooler thing is that, as with synthesis, arbitrary inputs can be given in, to guide the search, all in words!

6/8

It is quite impresive to see some of the responses - like Claude-3.7's analysis of some historical routes for strychnine!
@zlatkojoncev.bsky.social certified 👌

5/8

We made a small benchmark for this task **steerable synthesis planning**, with several targets and prompts, and the results are quite impresive!

Larger models are quite good at the task 🚀, with newer versions improving rapidly. A small model like gpt-4o-mini is not as good tho

4/8

In synthesis planning, we input mol. and the output are some routes, with no restrictions. But what if I want to specify details? - see img
In our work, users can specify a query, and during search the LLM selects only aligned solutions - working as a zero-shot value function!
3/8

Many tasks in chemistry can be tackled through search, like synthesis planning, mechanism determination, molecular design, etc...
Current methods rely on designed or learned value functions, but what if LLMs already know all of this? and how can we exploit it?

2/8

Special thanks to all the organizers, @benblaiszik.bsky.social @kj-schmidt.bsky.social @kjablonka.com @pepe-marquez.bsky.social @defnecirci.bsky.social for this amazing event and for the incredible effort in making this paper a reality!

Thanks also to @pschwllr.bsky.social and Victor Sabanza Gil with whom we organized the whole on-site event 🩷 🚀

To all participants, congratulations on your hard work and achievements!

Shoutout to @mlederbauer.bsky.social and @pschwllr.bsky.social with #GlossaGen
Jan Weinreich with #GeometricGeniuses
and the big team including Anna Borisova Marcel Moran Calderon Mark Tropin Rob Mills Yi ZHANG Stefan Peters and myself @andresbran.bsky.social for #MCPeptide.

Out of all the groups that we hosted in #Lausanne, **3** of them won prices with their projects (out of 34 amazing projects!), an incredible feat showing the huge talent pool we have over here!

Back in May, we hosted an on-site hub here in #Lausanne at @epfl-ai-center.bsky.social, generously supported by #LIAC (the @pschwllr.bsky.social group), and our sponsors Atinary, EPFL AI Center and El Tony Mate. Thank you very much Loïc Roch and Hermann Tribukait for your invaluable support 🫶!