Having gpt-5.5 iterate on models, giving it a target and letting it go works shockingly well.
I pointed it at labeled data, gave it a rough idea of the architecture I wanted and basically said “don’t stop until you hit X metrics” and it’s been chugging along making progress.
@jameygannon There is a big difference between working 7 days for yourself and working 7 days for someone else reaping the rewards.
I’ll work 997 for me and love it. Doing that for someone else? Not so much.
@pipuffin@lapislagoons You want the Nikon F2 and the nikkor 35 f2.0 or 50mm lens.
IMO there is no better camera made in the last century and it’s easily obtainable for $100 as a kit on eBay.
I’ve been working on this a lot in my acoustics project.
A few things that dramatically improved it:
- forbidding allocations, exceptions, and running with memory checkers that panic in those cases
- micro benches for everything
- break into small libs
- tests for everything
- clang-tidy rules that are very aggressive
With this set up, I don’t need to really look at internal code that’s generated, I can focus on the api boundaries and it’s generally good enough.
@usr_bin_roygbiv I’ve been mainlining their discography since the new album as well.
I love how even for tracks you’ve heard 100x you can find new elements to them.
@ashleybchae@beffjezos When people say that I add them to a mental list of people I keep to ignore.
Lots of things get you on the “pay no mind” list, but that’s a big one.
@mickeyxfriedman This goes with the “escape the permanent underclass” people who are like “you have 5 years to get rich, artistic or hot”.
I kinda agree with it a little bit, I’ve certainly been looksmaxxing more in the last year or so.
Learning too much. Just make the game, ship it, make another.
They’ll be terrible but that’s not the point. You need to learn how to finish games. You can’t learn it by reading about it.