querez

I didn't know about the static cache, that was a nice learning for me. The rest is pretty obvious /too lengthy for everyone who ever did profiling/benching. Also, the text has a lot of LLM slop-phrases in it that need cleaning up ("the X is structural", "the Y is real", "the gap is Z")

AI doesn't need to be AGI to be useful. Surely, you've tried Claude Code before and found it more helpful than Clippy?

> The project owner strongly emphasize the no LLM dependency, in a world of AI slope this is so refreshing.

The project owner is talking about LLVM,a compiler toolkit, not an LLM.

That's not how I remember it. Excitement for python strongly predated ML and data science. I remember python being the cool new language in 1997 when I was still in high school. Python 2.4 was already out, and O'Reilly had put several books kn the topic already it. Python was known as this almost pseudo code like language thst used indentation for blocking. MIT was considering switching to it for its introductory classes. It was definitely already hyped back then -- which led to U Toronto picking it for its first ML projects that eventually everyone adopted when deep learning got started.

I think you're underestimating what AI can do in the coding space. It is an extreme paradigm shift. It's not like "we wrote C, but now we switch to C++, so now we think in objects and templates". It's closer to the shift from assembly to a higher level language. Your goal is still the same. But suddenly you're working in a completely newer level of abstraction where a lot of the manual work that used to be your main concern is suddenly automated away.

If you never tried Claude Code, give it s try. It's very easy to get I to. And you'll soon see how powerful it is.

I assume TPU TCO is significantly cheaper than GPU TCO. At the same time, I also assume that market demand for GPUs is higher than TPUs (external tooling is just more suited to GPU -- e.g. I'm not sure what the Pytorch-on-TPU story is these days, but I'd be astounded if it's on par with their GPU support). So moving all your internal teams to TPUs means that all the GPUs can be allocated to GCP.

> No Gemini was not "entirely trained on TPUs". They did hundreds of experiments on GPUs to get to the final training run done entirely on TPUs. GCP literally has millions of GPUs and you bet your ass that the gemini team has access to them and uses them daily.

You are wrong. Gemini was definitely trained entirely on TPU. Of course your point of "you need to count failed experiments, too". Is correct. But you seem to have misconceptions around how deepmind operates and what infra it possess. Deepmind (or barely any of Google internal stuff) runs on Borg, an internal cloud system, which is completely separate (and different) from gcp. Deepmind does not have access to any meaningful gcp resources. And Borg barely has any GPUs. At the time I left deepmind, the amount of tpu compute available was probably 1000x to 10000x larger than the amount of gpu compute. You would never even think of seriously using GPUs for neural net training, it's too limited (in terms of available compute) and expensive (in terms of internal resource allocation units), and frankly less well supported by internal tooling than tpu. Even for small, short experiments, you would always use TPUs.

I grew up in the Internet at that time, and it's certainly not how I type. So you might want to be more specific about which sites or subcultures you think this style is representative of?

The sentence is constructed, weirdly, but it's meant to say that fever is "killing off unknown foreign bodies"

Arguably both Go and Python also have great stdlibs. The only advantage that JVM and .NET have is a default GUI package. Which is fair, but keeps getting less and less relevant as people rely more on web UIs.

I don't think "inference" as a verb has become technical slang. At least not in my bubble.

"developers can prototype, fine-tune, and inference [AI models]"...

shouldn't it be infer?

I think what is distinct in this proposal is that there are n 1:n channels

Why? I fail to see how using chromium as basis for other apps has impact on who has the power to innovate in the browser space?

> The geometric mean of the timings improved from 218 to 12, a ca. 20× improvement.

Why do they use the geometric mean to average execution times?