@ArielVinda@rfleury@typesfast if we are talking anecdotes, I've interviewed, hired and fired more than a hundred dev's in my career. I've reviewed a ton of code as a CTO and lead dev. current LLM's don't produce what feels like mean human code to me. that's why the argument feels flawed
@rfleury@typesfast isn't the code that type checks, compiles, and passes tests (specification validation) meaningfully better than the one that doesn't? I don't want to waste your valuable time Ryan, I love your work. just this particular point doesn't feel right
@rfleury@typesfast this is true for RLHF, not RL in general. they are using compiler errors, test results, benchmarks etc as RL signal. so it's not merely ~mean programmers reviewing it
@rfleury@typesfast small correction: it doesn't "consider ~mean quality output as successful by definition" because of reinforcement learning. with RL whatever produces better reward signal - passing tests, lints, etc. is reinforced. so the output is better than mean
@vkrajacic Totally agree, but I think it won't matter soon, LLM's will fully reverse engineer anything from binary alone. Even SaaS can be replicated from input output. What do you think is a correct strategy long term for indie software?
@XorDev If intelligence provides the direction, will provides the magnitude. It's better to move slowly in the right direction that fast in the wrong one.
@figwit@dhh Interesting, so I lucked out with 5090. Honestly I was surprised when it suddenly started working. And it continues to work without issues.
@figwit@dhh both. the driver is not pushing the frame to DP on 3070 even though it works on Windows but the same driver is pushing it on 5090. I have little hope they'll update the driver to wok on 3070
@CaptainGPU Може ~1МБ за 5 хв. Instruments каже що leak-free. Можливо фрагментація пам'яті в аллокаторі росте, тобто вона звільнена, але аллокатор не може повернути сторінку пам'яті до системи.
@argumatronic Your books and blogs helped me a lot with making sense of Haskell, abstract algebra and category theory as a complete beginner. Thank you for all your work!
RELEASE DAY
After almost 10 years of hard work, tireless research, and a dive deep into the kernels of computer science, I finally realized a dream: running a high-level language on GPUs. And I'm giving it to the world!
Bend compiles modern programming features, including:
- Lambdas with full closure support
- Unrestricted recursion and loops
- Fast object allocations of all kinds
- Folds, ADTs, continuations and much more
To HVM2, a new runtime capable of spreading that workload across 1000's of cores, in a thread-safe, low-overhead fashion. As a result, we finally have a true high-level language that runs natively on GPUs!
Here's a quick demo:
@ChShersh Just ordered mine a week ago (used Microsoft sculpt before). Now I'm scared. Why it's such a painful transition for you?
Have you tried The Platform too?