Viewing profile — wren6991

I'm fine with a 1k-line function if you just have that many things to do in a row without taking a breath. Breaking it up into smaller functions feels neater when you write it, but…

Sorry if the joke didn't land; I have heard a lot of different numbers for the size of US labs' models, but never seen any of them substantiated, so I think you're likely to just g…

In this situation it's completely moral to go pirate the games on a different platform. Illegal, but moral

I'll ask my uncle (he works for Nintendo) and get back to you on that one

Kimi-K3: 2.8T Qwen3.8-Max: 2.4T DeepSeek V4 Pro: 1.6T DeepSeek V4 Flash: 284B (all are total parameter counts, not active parameters)

"Now I have the full picture. Let me just undo that last change." $ git checkout . "Oh no, that undid everything."

Another banger from the mind that brought us Beads, the solution to having your repositories littered with Markdown files, written in a svelte 250,000 lines of Go. What's that? The…

I guess they are saying sizeof(...) should return the annotated static size rather than sizeof(int*) in this case

Opus 5 clearly frogmaxxed. gemini-3.6-flash runs 2 and 3 responded best to the royal portrait context.

> (everything here also applies to c2y's newly introduced _Countof) Stop the press, they're adding what ? So I don't have to keep adding #define count_of(arr) (sizeof(arr) / sizeof…

Trying OpenRouter gave me a strong impression that LLMs are already a commodity and individual providers are not that important. Providers don't seem to be too happy about it.

As a frog I'm happy to report the water is comfortably warm today, the same as it was yesterday.

Huh, so every interactive event has a network request in the loop, with perceptible latency? And you still end up obligating client-side JS for the plumbing? I'd been meaning to lo…

I don't think so. Probability of what? If you put something through a softmax the output is (trivially) a valid PMF. Does that matter? You're not sampling from it.

The worst part of this workflow is when you accidentally run `git commit -am` at the end and destroy all of your careful staging.

This seems functionally similar to OpenAI having a step in pricing once you exceed a certain context length (also at 272k aka 2^18 aka 256k). Having a lot of active context increas…

I'd guess because it costs them nothing and it gives you a smoother transition back towards paying for their products.

Ooh. This looks like an interesting paper and there were a couple of things in the intro that I found counter-intuitive. It'll take me a while to digest the whole thing. > Coverage…

That's true. The fact that an LLM is a pure function of (all previous tokens) -> (next token), with internal state like KV cache only existing for optimisation purposes, is pretty …

The "next-token predictor" framing is also a bit shaky. It's an accurate description of pre-training, where next-token prediction is a useful learning objective to force the model …

Sounds a lot like running the Qwen3.5/3.6-series models at home: you need checkpoints for the recurrent state (GDN in the case of Qwen). You avoid the miss for the common case of 1…

> Anthropic has never advocated for a ban on open-weights models. What are the legal ramifications of this statement if it turns out Anthropic have lobbied for this? Does it just g…

Older models did show the full unredacted thinking trace, but I don't think Opus has ever shown full CoT. Here is an archived version of Anthropic's API docs saying that Sonnet 3.7…

The main source I'm aware of is this Reuters article: https://www.reuters.com/world/beijing-is-looking-curbing-ove... However you should take it with a pinch of salt because IMO th…

I think the type of persistence rewarded by benchmarks may be misaligned with instruction following