ray__

Protein binders (as in, small proteins like the one in the article that themselves bind other proteins) aren't a rate-limiting step, but also are fairly restricted in their applications as therapeutics. Designing small-molecule ligands is much more challenging, and AI/ML design tools for small molecules are much less effective for a number of reasons, including 1) existing training data are sparse the search space for small molecules is arguably much larger and more complex than for proteins 2) the evolutionary data that powers protein structure prediction and design don't exist for small molecules. So it is still non-trivial to design a small-molecule binder for a protein.

It's true that clinical trials are a big cost center for drug development, but there is an argument for mitigating failure here earlier in the process through better technology development. Some of this will come from AI and some of it from better experimental methods for probing biology. One of my favorite efforts in this area is OpenADMET, which combines ML and experimental methods for addressing tox-related issues in drug discovery, which are common causes for failure in the later (more expensive) clinical trial stages:

https://www.nature.com/articles/s41467-026-73410-8 https://openadmet.org

Every time I get blinded by this I wonder if it's really necessary at all...I can turn the brightness up myself if the QR code doesn't scan.

Anecdotal, but I tried running a few identical biology questions through both Fable and Opus and the classifier was only rejected my queries with Fable.

Any previewers have hot takes? I've really preferred gpt-5.5 over Opus 4.8 for data analysis and scientific software work. It seems much more reliable. Fable is unusable for the type of work that I do (due to guardrails). Really looking forward to trying these new OpenAI models out.

Great article. This is the kind of web design that I like.

I see, very interesting, thanks!

This is a cool idea—I know from snooping on sumbit scripts and node utilization on the HPC that I use at my institution that most submissions leave some compute on the table (and many of them are egregiously bad). I'd probably vote in favor of sending every submitted sbatch script through an LLM (at least for everyone else, I'd would prefer tuning my own usage myself :) ).

Presumably the underlying model here is also an LLM? To what degree is it "fine-tuned", or is it just given a set of tools to build a good picture of cluster usage?

I wonder if the tendency to write short punchy sentences stems from deliberate RL efforts to avoid repetitive, consistent writing? I seem to remember that a critique of early LLMs was that they would produce sentences whose construction was too homogeneous. Would be interesting to know the answer to this.

You could say something similar about biology—just physics behind the curtains, and we understand a lot of the basics. The difficulty comes from complexity, not mysticism.

To be clear I don't think that LLMs are sentient, but the appeal in studying them is similar to biology in that you get to dissect a highly complex system with comparatively crude tools.