Andrew McCalip
@andrewmccalip • 79,427 subscribers
https://t.co/XTkNjjVHci Building spacecraft @vardaspace. Circumnavigating world @ https://t.co/hLM2DEGTxZ. Former: Co-Founder Cosine Additive, acquired by GE
Shorts
Videos
Immaculate vibes from the deck of Project Bob So far the score is: autonomous boats: 0 ocean: 3
Andrew McCalip132,719 просмотров • 1 месяц назад
Took the morning to mess around with Opus 4.6. My go-to model eval is always the same. I ask for a one-shot, browser-based STEP viewer plus a tolerancing/markup app. Took me two coffees to get it running. This is what I constantly beg every US-based online manufacturer to offer. I think it’ll be a game changer for the industry. Everyone hates parsing paper drawings, so we default to blanket .005 tolerances. Go a tiny bit further and let users tag a few faces. There’s a nugget of the future in this approach. Probably half stupid. Definitely inappropriate for some parts. I know, I know. I deal with real aerospace-grade drawings every day. My hands still wet with coolant as I type this. Still has the taste of the future. Anyway, I digress. Basically a PDF equivalent for CAD. Turn a STEP file into a traveling source of truth by layering metadata onto face IDs. Colors, threads, tolerances, notes, whatever. The model isn’t the CAD file anymore. The STEP plus its annotations are. Of course it’s never actually one-shot. It’s an hour of back and forth. Tight loop. Tiny spec changes. “No, not like that, like this.” I run the same test on every new GPT/Claude release. I’m still not sure I’m good at programming. I just know how to ask for things clearly, write specs that don’t leak ambiguity, and iterate like a maniac until the thing works. Manifest the idea into existence, then it becomes the spec for the next iteration. Humans are awful at defining requirements upfront. You manifest the idea into something real, then that artifact becomes the spec for the next loop. This feels like the primary skill of the future. It’s project management, but hyper-accelerated. Micro-management. Pico-management. An iteration speed you could never ethically apply to humans. And I’m slowly realizing the harness isn’t for the model. It’s for me. I’m the bottleneck. The model can sprint forever. I’m the one that runs out of clarity. Anyway, this is a throwaway idea. Maybe some YC kids will raise on it. Whatever. Ideas are cheap. This took an hour. Repo link below. Have fun.
Andrew McCalip161,310 просмотров • 4 месяцев назад
Turns out the UI couldn't parse the negative RPM value to slow it down 😂
Andrew McCalip508,700 просмотров • 1 год назад
Больше нет контента для загрузки