 | TL;DR: A Fireworks AI cofounder called something most AI-wrapped founders are quietly standing on "insane" — and it's not the technology.
It's the lifeline. Benny Chen tunes open-source models for real companies — Cursor, Cognition, Harvey — and his actual argument is simple: if one frontier lab's pricing or policy call can break your business, you can't honestly value that business ten years out, no matter how good this quarter looks. His fix isn't building a frontier model from scratch. It's tuning an open base to your own niche, then building the data flywheel that keeps improving it — so the base model underneath you can be swapped out entirely and the business doesn't blink. Same week this episode dropped, Thomson Reuters backed the identical bet with real money — building its own model off an open base specifically so it "fully controls" it instead of renting someone else's roadmap. Different scale, same math.
My friend works in an electronics factory, stationed in the Penang Island industrial zone. He told me his factory and all others like his in the zone are essentially in a whirlwind of "red ocean" – fighting for price competitiveness. He says, all the valuable part of the upstream work were already taken up by other countries, such as China and Vietnam. And only the very low-value works get trickled down to Malaysia. The factory doesn't own any intellectual property developed from innovation. They just do whatever was handed down from the upstream industries of other countries, lightly process them, package them together, and ship them overseas. Hardly anything groundbreaking. And so, they are always susceptible to external factors, such as US tariffs, etc. US already marked our countries as one of the suspected offenders of 洗产地 (origin laundering). And so, the tariffs really weigh us down, and reducing our overall profit margin. Not that it has anything to do with my friend. He just works there - getting piled on with additional work under unrealistic expectation, in a department going through high-turnover of staff, and without a clear path forward. But he still makes the best of what it is. And what it is - is figuratively anchoring on shifting sand in a stormy weather, rather than on solid rock. ___________________________ The same fault line keeps showing up once you strip the industry names off: whoever owns the base layer sets the terms, and everyone standing on it finds that out eventually — a founder renting a model, a factory renting someone else's spec sheet, doesn't matter which.
Actually, this reminds me of something — Guillermo Rauch (Vercel's own founder, one of the companies Benny's team tunes models for) laid out almost the exact same fracture from the engineering side a few weeks back: the role itself is splitting into an execution layer and a judgment layer, and only one of those is safe to not own.
Are you building on ground you actually own, or ground you're renting? Drop your take.
Clip credit: Inside the Silicon Mind with Firas Sozan — full video on their channel. DM for credit or removal requests. [link] [comments] |
Predictive technology: computational social science
