I built an eval to measure the ability of various LLMs to create CAD. Gemini 2.5 and o3 are leading the pack.
It looks like reasoning may be a big enabler. Anecdotally, I've noticed that the reasoning models effectively think through how they're going to make the part.
It's crazy how great LMs are at 3D modeling, a task that was probably a very small % of their pre-training and probably 0% of their post-training.
From @wgpatrick: https://t.co/4hm4gYCE6L
New blog post.
Vibe Code Everything.
I've found it stunningly useful to use Cursor to do non-software tasks. For example, creating a book for my son and creating a financial model. Other founders I know are doing the same thing.
https://t.co/lObCyUemcU
@julianlehr But isn't telling a LLM, "Make me a nicely formatted PDF of [whatever the thing I'm working on with the LLM", a form of hyper data compression. You're effectively replacing 100s to thousands of line item actions ("Ctrl-B for Bold") with a simple command.
@SynBio1 I think that all parts of the infrastructure to develop or make industrial products produced will remain, by and large, centralized.
One can imagine a maker community that makes real products. But - it will probably remain small compared to industrial applications.
@alexlorestani In other words, the capacity apocalypse is destiny, if there are a number of successful molecules hitting the existing CMO fermentation capacity at the same time. There would be a crunch; that would be resolved once supply catches up. But it will take some time.
@alexlorestani With a successful product-market fit and good economics (and, thus, to your point, interest from CMOs) -- the challenge, I imagine, becomes the timeline to build out additional capacity for the CMO. Usually, it's 3-5 years.
@alexlorestani If (big if) alternative protein cost per kg gets to where we all want and hope it to be (<$10/kg), then you're totally right. I've done some very basic modeling that estimates the infrastructure need to be ~10BL, something like 100-200X today's global capacity.
15 minute pressure altimeter change via ASOS NWS/MADIS 5 minute interval data. Shows the shockwave from the #Tongaeruption , feel free to use as you wish.
Can DNA be made to assemble into giant structures as intricate as living cells? In the preprint I led, we put ~1000 unique origami nanorods into shapes like this ghost, with a footprint larger than a single E. coli! (https://t.co/afKdSKB7JQ)
@willcanine@opentrons I hope you feel great and proud of what you built. Opentrons has made a huge impact on the lives of many scientists and non trivially democratized biotech tools. Looking forward to seeing what you don next and hope to hang soon!
We'll be using this new capital to build out larger-scale biomanufacturing infrastructure, such that our customers can run pilot and full-scale manufacturing runs.
We at Culture are thrilled to announce our $80M Series B financing led by @npv_vc and @Synthesis_Cap. We're so excited for whatβs to come as we tackle client demand for large-scale biomanufacturing. Thanks to @Forbes and @johncumbers for sharing our story!
https://t.co/mnC3GxvEO6