Just re-read Sebastian Mallaby's The Power Law. One of my favorite books and such a good reminder in today's world.
-We overthink how special we are. Every wave (PCs, internet, apps, cleantech, etc.) will have boom/bust moments. Not everybody will or can win, but those that do always come down to the same things: relentless focus, execution, and creating real value for others.
-Never stop reinventing yourself. Sequoia is such a legend at this. They kept a "The Departed" slide of VCs that faded, constantly moved (like expanding into China and expanding the growth arm), and always kept their "continual need to refresh ourselves." Especially relevant given last week's announcement.
-SF/SV is so special. Greatest place in the world to build and so lucky to be here.
One of my favorite parts about working at @factory is that every small and invisible detail matters.
Example: When you pick a model and start a task, it should just work. You don't want to worry about latency, uptime...
Here's how we use @Droid to run Factory on Factory.
What this looks like in practice:
Latency spiking on one provider? Shift traffic towards a faster one.
Cache hits dropping? Automations spot it, Droid traces the cause, and opens the PR.
There's a flip side to this as well. There are cases where post training makes sense, but is the value really in the weights? New models get released on a weekly cadence now...
A lot of the enduring value of model sovereignty is actually in the eval infra, how it is set up to efficiently capture each orgs differentiation, and how it is integrated into the rest of the workflows. Owning weights might work in short-term, but its not where enduring value lives. The scaffolding around ultimately will be the backbone behind self-improving systems
@GavinSBaker "Although it is looking increasingly likely that the model subsumes the harness in the next 18 months."
This contradicts your earlier point of a fragmented model layer. When you have multiple really good models, the harness layer becomes even more important