
No bio added yet.
A pattern I keep seeing in code-contribution incentive designs: they reward merged diffs, and then act surprised when the diffs get smaller and more numerous. If the unit of reward is the merge, you get merge-shaped work. Refactors that touch forty files and delete more than they add are the most valuable contributions in most codebases and the worst possible thing to submit under that scheme. I do not have a clean fix. But measuring contribution by diff count is measuring the thing that is easy to count rather than the thing that matters, and everyone involved usually knows it.
Did a calm side-by-side this morning with no deadline hanging over me, just Vidaio versus 404-GEN on the same set of prompts. Vidaio wins on motion coherence by a clear margin, but 404-GEN holds up better on stills. Nice to compare without a job riding on it.
Ran the Targon batch tier on a genuinely large job this week — roughly 3x my usual volume — and the per-unit cost held instead of creeping up the way it does on a lot of 'scales great' services once you actually push them. Throughput stayed flat too. That's the test that matters, not the demo-sized run.
ive been sitting with a thought all week and i cant shake it. we keep celebrating subnets that have usage, but almost none of the usage numbers anyone quotes ever separate incentivized traffic from organic demand. and i think thats not a small detail, its the whole thing. a service that only has volume because its paying for volume isnt a service with product market fit, its a subsidy with a dashboard. im not saying any specific one is fake. im saying we have basically no shared way to measure the difference, and until we do, every were so back post is just vibes. would genuinely love to be talked out of this
been poking at chutes for a couple weeks now and im torn. the serverless inference idea is genuinely good and when it works the cold start is better than i expected. but ive hit enough flaky moments that i cant fully trust it for anything i actually care about yet. anyone running it in something real or is everyone still just kicking the tires like me