@businessbarista How are they solving the level 3 problem? I imagine most companies on their AI transformation journey should make this level 0, cos there’s no getting around it.
@njavwamutambo@ycombinator@snowmaker Also, there’s no reason to build your product with just one model, you should be leveraging multiple models and routing tasks and workflows to the most appropriate model.
@njavwamutambo@ycombinator@snowmaker I don’t see the connection between YC not funding African startups and these startups preferring Chinese models. It’s just based on cost, these models are cheaper, and provide good enough performance all things considered. Also, big names like Cursor also use them.
As a current masters student, I’ve tried 5 times today and I’ve been automatically rejected every. Single. time despite providing all the requested docs, and there’s no way to escalate to an actual human to help.
I’m a fan of automation as much as the next person, but we need to take a step back and ensure it’s working as intended. @github has an education benefits program for students and teachers, to qualify you have to submit documentation to validate you’re a current student/teacher
Built a solar forecasting app this week using AI coding tools. The speed is genuinely incredible — physics engine, multi-string support, calibration loop, public API. Things that would've taken weeks done in days.
But here's what I didn't expect: the QA burden.
So the bottleneck isn't capability. It's token economics. For AI-assisted development to expand meaningfully beyond SWE, tokens need to get better and cheaper — not just incrementally, but by an order of magnitude.
We're not there yet. But the direction is clear.
@0xa5ad@balajis yes and no. I agree that agents acting on behalf of users should be granted similar privileges, but w/caveats. things where the user has already paid, should remain intact. “free” often means ads, so we’ll need new monetization models in lieu of ads for user-directed AI agents