nbardy

6 ms·

31.
Show HN: I Made an Open Source Swarm IDE(nbardy.github.io)
2 points by nbardy 7mo ago|0 comments
32.
by nbardy 7mo ago
You can estimate on tok/second The Trillions of parameters claim is about the pretraining. It’s most efficient in pre training to train the biggest models possible. You get sample efficiency increase for each parameter increase. Howeve
33.
by nbardy 7mo ago
How much of your RAM does that use including kv cache. Is there enough left to run real dev workloads AND the llm? Also can you run batchwise effectively like vllm on cuda? Enough to run multiple agents at the same time with throughput?
34.
by nbardy 7mo ago
Why does apple want to make this hardware hard to access? What actual benefits do they get? I guess they can have their own models run faster than the competition on their hardware? But they don't even really have anything that consume
35.
by nbardy 8mo ago
They are far behind. Go check re-swe bench to see the overfitting measured Or just try to use them. They don’t generalize as well. They are benchmaxxed.
36.
by nbardy 9mo ago
They should probably fund their military first. It’s petulant the way the EU is throwing a hissy fit after we’ve had lop-sided trade deals for years and funding the entire NATO alliance ourselves. They act like we’re going to war with them
37.
by nbardy 9mo ago
No it’s not. I have written cuda kernels and 8bit optimizers with this. They’re actually very good at speed optimization and can iterate very quickly taking notes on trials and failures and benchmarks. I’ve had it write 10 different attempt
38.
by nbardy 9mo ago
> Claude Code officially added native support for the Language Server Protocol (LSP) in version 2.0.74, released in December 2025. I think from training it's still biased towards simple tooling. But also, there is real power to simp
39.
by nbardy 10mo ago
Your way off, this reads more like anti capitalist political rhetoric than real reasoning. Look at Nvidia nemotron series. They hav become a leading open source training lab themselves and they’re releasing the best training data, training
40.
by nbardy 10mo ago
When are people going to drop the immigration is good at all costs assumption. We need a well managed set of immigration polices or country WILL take advantage of US. These are our military rivals and we sell our most advanced math, physics
41.
by nbardy 10mo ago
Those arc agi 2 improvements are insane. Thats especially encouraging to me because those are all about generalization. 5 and 5.1 both felt overfit and would break down and be stubborn when you got them outside their lane. As opposed to Opu
42.
by nbardy 10mo ago
You haven’t actually looked at their fundamentals. They’re profitable serving current models including training costs and are only losing money on future RD training, but if you project future revenue growth on future generations of models
43.
by nbardy 10mo ago
Why are you assuming Anthropic is for sale? They have a clear path to profitability, booming growth, and a massive and mission driven founding team. They could make more money keeping control of the company and have control.
44.
by nbardy 10mo ago
This is misleading. They had 4.5 which was a new scaled up training run. It was a huge model and only served to pro users, but the biggest models are always used as teacher models for smaller models. Thats how you do distillation. It would
45.
by nbardy 11mo ago
There’s another paper that shows you can get the same effect by training auto regression on Fill in the middle data. So it’s more about the mask modeling objective than Diffusion.
46.
by nbardy 11mo ago
This is so not a problem. After a long stretch of I just ask codex to: Audit the code base and event changes and describe the data model and all related and possibly overlapping functions. Plan a new redesign that is simpler, has less code,
47.
by nbardy 11mo ago
They have been able to write languages for two years now. I think I was the first to write an LLM language and first to use LLMs to write a language with this project. (Right at ChatGPT launch, gpt-3.5 https://github.com/nba
48.
by nbardy 11mo ago
Yea, I can't get gemini to stop and think, even if I tell it to not write code it will rewrite the code block each time
49.
by nbardy 11mo ago
This isn’t gaming the benchmark though. If training on similar data generalizes that’s called learning. Training on the exact set is memorization. There is for a fact teams creating puzzles to RL against as training environments. As it’s be
50.
by nbardy 11mo ago
The techniques here are 100% transferable. It would take some work to migrate it to diffusion + images. But if you tuned the input prompt and rejection detector that is fairly trivial work in a few days.
51.
by nbardy 11mo ago
No they can't, it's a N^2 algorithm, just fitting it in the context window is a challenge. And sure maybe not 2mil of it is usable, but they're reliably pushing the frontier here.
52.
by nbardy 1y ago
You know you can AI review the PR too, don't be such a curmudgeon. I have PR's at work I and coworkers fully AI generated and fully AI review. And
53.
by nbardy 1y ago
The serving infrastructure becomes very efficient when serving requests in parallel. Look at VLLM. It's the top open source version of this. But the idea is you can service 5000 or so people in parallel. You get about 1.5-2x slowdown o
54.
by nbardy 1y ago
These estimates are way off. The concurrent requests are near free with the right serving infrastructure. The throughput per token per dollar is 1/100-1/1000 the price for a full saturated node.
55.
by nbardy 1y ago
The local employees would be able to negotiate higher wages given the absent It provides supply in the market and lowers their bargaining power, so the parity in wages is parity in a lower baseline Also in practice there are salary ranges a
56.
by nbardy 1y ago
I love the symbol LLM first approaches. I built a version of this a few years ago as a LISP https://github.com/nbardy/SynesthesiaLisp
57.
by nbardy 1y ago
I bet you can close the gap with a finetune. Should be quiet easy if you have some o4-mini results sitting around.
58.
by nbardy 1y ago
I hate to be pedantic, but the llm is definitely doing embedding math. In fact that’s all it does.
59.
by nbardy 2y ago
This is untrue. You can be over specified in your prompts and say exactly what types and algorithms you want if you’re opinionated. I often write giant page long specs to get exactly the code I want. It’s only 2x as fast as coding, but thin
60.
by nbardy 2y ago
The market is betting the best model wins to some extent and anthropic is not slowing down.

More ›