Most AI labs publish two kinds of writing: papers, and press releases. We think there's a third kind worth doing — the working note. Shorter than a paper, more honest than a press release.
What this blog is for
Three things, roughly in order of how often you'll see them:
- Research notes. Partial results, negative results, and methods that
worked. If we measure something interesting about how models reason, it goes here first.
- Engineering notes. What we learned building the substrate — the
practical decisions behind provenance, calibration, and stability.
- Announcements. Alpha access, papers, hiring. The rare posts.
What it isn't for
Hype. The field has enough of it. If a post here makes a claim, the claim will come with the evidence that produced it — that's the whole thesis of the lab, applied to its own writing.
We will not ship systems that confabulate. The same rule applies to blog posts.
If you want these in your reader, there's an RSS feed. The first substantive research note lands alongside the closed alpha.