Launching the wetlab!

The biggest unlock of the week is that we now have a wetlab! We signed our lab lease with next Monday, September the 21st, as our first day. On the one hand, while waiting for it, we’ve had enough time on the AI side to develop a working prototype of our scientific engine. On the other hand, I feel excited and anxious about the influx of wetlab challenges. Luckily, the first chunk of investor money in our bank account adds serenity.

We’ve talked to quite a few people by now — there’s a lot of interest on all axes. We’re trying to figure out how to balance building with fundraising on the fly. I believe the pitch got stronger after a few rounds of feedback and improvement. Consciously, we were initially trying to go as broad as possible in pitching, while focusing short term on a very specific path to a tangible PoC on the execution side. It has bitten us a few times, and now we have bridged the gap closer. You can check out the second iteration on our strategy with a more pragmatic scope.

We’ve sketched out a ballpark for our PoC budget (from the simplest tier with a single cracked problem to the most challenging one where we would target 4-5 orthogonal problems in parallel).

On the AI side, we got excited about the recent twitter discussions on the novelty of AI discoveries (Navier-Stokes), and unleashed our AI engine to scout the prior art out for a bunch of recent claims about AI making a scientific discovery. “A month in the lab saves a day in the library”, or a questionable claim may trigger a lot of backlash, if one hasn’t done their work on figuring out what actually has been already done. Enjoy!

I wonder if the frontier labs have overindexed on sales and PR — their current tier of frontier models is definitely bleeding tokens, and works extremely slow, while I wouldn’t say that there’s a sensible improvement in quality for the work I’m using them for. For most of the tasks, open models are good enough, and usually way faster and cheaper.

Maybe, we may be overstretching ourselves a bit — but it’s definitely super fun! I do like that we iterate fast, learn from reality and improve. And, the feedback from reality has been overall strongly positive, which is inspiring!

Share on: