kristjansson

Because all the constitutional arguments therein are a long-shot attempt to remove a long-litigated fine by attacking the underpinnings of the administrative agency that that assessed the fine. It&#x…

Explicitly considered. The ultimate fine is much _less_ than it could have been under the controlling law.

They had several opportunities to appeal, which they took advantage of, and which _reduced_ their liability at each step

This headline and article seem like an over-reading of a technical footnote in the cited case. The relevant part of the footnote (citations omitted, emphasis added) > It’s not clear how equating “…

Congress not only _could_ control rents nationally, congress _has_ controlled rents nationally [1] and those controls survived the review of the court. [1]: https://en.wikipedia.org/wi…

Trading currencies is basically zero-sum. A VEF/USD seller needs a USD/VEF buyer. That buyer demands some VEF for each USD i.e. the floating exchange rate. In a free market, imbalance in t…

> Tbill interest or similar. They have $100bn dollars and their only job is not to lose any of it.

The ATC union has some unique history [1] [1]: https://en.wikipedia.org/wiki/1981_Professional_Air_Traffic_... …

We’ve built good enough systems that lots of people have never had engage with the complexity of reality?

The happy path is easy, as you imagine. It’s all the other imaginable and unimaginable paths, each unhappy in their own way… Oh and thousands of lives hang in the literal balance

Positive Train Control has been a big fight too.

> Bachelors in Aviation and Aerospace Science with an Air Traffic Control Concentration I’m not that far removed from my time in college, but I’m shocked by the specificity of that.

Replaced, no. ATC, at least while we still have human pilots, is a system for instructing, organizing, and responding to humans, with all their flexibility and foible. Now a copilot e.g. ATC audio pa…

Oh, algorithms, they should have thought of those.

Too low, most likely.

So you wouldn't want to be under it, but a good-sized hill would be enough cover?

If your data is too sensitive for AWS, you're in a different realm that most enterprise users.

Most people really do not know what they want at any level of detail.

Totally, they did great work under their constraints. Training in FP8, the MLA thing they introduce in DeepSeek-V2, etc. I just take particular issue with the attention the PTX thing is getting beca…

Per the technical report: > The minimum deployment unit of the decoding stage consists of 40 nodes with 320 GPUs. but realistically, >=671GB of VRAM to run at full precision on GPU, or >=131G…

To be sure, DeepSeek did great work, and this is a bit aside from TFA. But the PTX thing is a bit of meme? What do we think torch.compile and triton and llvm's nvptx backend are doing under the …

Ollama is doing the community a serious disservice by presenting the various distillations of R1 as different versions of the same model. They're good improvements on their base models (on reason…

Only because OpenAI overpriced o1, like they did with GPT-4.

Writing a few intrinsics where necessary is not really comparable to the work required do reimplement something like CUDA on AMD (or equivalently deal with ROCm)

A careful implementation adapted to a specific workload is always going to beat a more generic implementation.

This whole episode is weird. I can’t tell how much of the popular reporting is misinformed and how much has been disinformed. R1 (and sorta V3) are clearly progress, but are definitely not step-functi…

This is ridiculous. Since the actual training code for DeepSeek is _not_ public, this is a based only on the technical report, which mentions PTX one (1) time in §3.2.2 Efficient Implementation of Cr…

How did the deficit trend 2017-2019?

closer than $0.05

Clock time without a time zone is underspecified. It might refer to the users current timezone? Or at the time of the event? But what if you invite a user in another timezone? You’re going to miss eac…