qrios

2025

I had my first experience with spreadsheets in the early ’90s using Improv (in Hanover, home of CeBIT, there were lots of exotic computers available cheaply on the “second-hand” market).

Thanks to my experience with Improv, I later found it extremely easy to work with multidimensional data. In fact, I recently built an in-browser version of it. (Regardless of that, I highly recommend that everyone take a look at “PivotTable.js” [1]. Essentially, a direct implementation of the Lotus Improv UI.

[1] https://pivottable.js.org/examples/

„MacBeth“ like „MacPro“ or like „BigMac“?

Not related to „atmosphere“ at all.

„Atmosphere Explorer“ like in AT Protocol Browser …

From Dune by Frank Herbert. See "Butlerian Jihad" [1]

[1] https://dune.fandom.com/wiki/Butlerian_Jihad

> I had thought Xilinx ISE and Altera Quartus had the market cornered on the most difficult environments to get running, …

These tools were the reason why I gave up on FPGAs years ago and chose to spend my time on GPUs instead. And to this day, I still don't understand why they worked so differently from what I had expected in so many ways.

Contend of the LLM calls.

> Rings forged from meteorites

When I first read the title, it sounded to me like a classic case of Kessler syndrome.

Fun story, thnx for sharing!

The most amazing thing about it is probably not the “upload” itself, but the lossless compression ratio of approximately 150,000.

> "… outright revolt begins within another 100 subjective hours."

No wonder: the brain has to work a third of the time, with no weekends and no beer on Friday nights.

„Published July 2026“

Even a month can change a lot these times.

Works better than in Claude Code.

From the README.md: If "fancy new AI tech in an old-school minimalist package" sounds like your vibe, you might like this.

What is the question about:

A: Why start a 'new' software project? (instead of old, none, multiple, ...) B: Why in 'C'? (instead of Mojo, Java, D, ...) C: Why 'today'? (instead of Yesterday, Tomorrow, never, ...)

For me the beauty of this project is:

Someone tried to figure out for themselves what an end-to-end process would look like for what they wanted to do. What happens to the input? How are files written? What is sent to the LLMs?

And they used the tool they know best. No `import openai`.

The interesting part is what they try to not to say: More indications for K3 is based on distillation from Claude and GPT.

From [1]:

> As you might guess, this suggests that distilling reasoning traces may have been possible for a long time without ever breaking the cryptography.

> An anecdote: we find that prefilling Kimi-K3 reasoning with a few tokens of Opus reasoning measurably shifts its response toward Opus’s

> A small memorization analysis showed that specific Claude and GPT reasoning spans are up to ~6 orders of magnitude easier to extract from Kimi-K3 than from the next-closest model.

[1] https://x.com/kotekjedi_ml/status/2087147042888114428?s=42

I'm sure this will become the standard. And plastic is an excellent analogy. Maybe we can take it a step further and compare it to on-demand 3D printing.

Why would anyone still use off-the-shelf software when they can have a system that has access to all data, can transform it into any form, and can export it in any format?

After years of thinking that I needed to develop a decent movie management system for my own films or a columnar browser for large CSV files, Claude and Qwen each delivered exactly what I needed in just a day.

The term “currently” is the sticking point here. As far as I know, there is still no one who can provide evidence as to why it should remain that way. And the "by luck with RL" is soon not by luck anymore. The more stupid and not so stupid ideas are discussed with the LLMs, the chances are growing that it is not a question about luck but by growing probability.

> FWIW, China is suddenly talking about model export controls. ...

Thanks for the hint. It makes sense, even if the main driver for releasing the models is to reduce the market size for the US companies.

Do you have any sources?

Edit: "market size"

As someone who usually works with data analysis, the distribution of the numbers strikes me as odd. Almost all of them have one number that appears four times, and one or two that appear three times. And overall, there are an unusually small number (0–9) of digits that appear at all.

Maybe it's not just vibe-coded, maybe the numbers themselves are being hallucinated by an LLM.

A new CUDA API? For DOS?

Thanks for sharing! I had no idea about the “3-30-300 test,” even though I always pay attention to city trees.

I can, however, easily explain the division in Europe: In Italy (for example, in Palermo), the vigorous growth of many species very often leads to significant damage to infrastructure.

Here in Vienna, there’s a directory of trees[1] where you can see, among other things, the species and age.

[1] https://baumkarte.at/

> burning tokens like there was no tomorrow

Literaly.

Another „great filter“: How to handle dagerous information?

We are using it for "old-fashioned" use cases (sentiment, classification) for some clients here in Europe. Mistral Small 3.2 8bit is good enough for most well-defined cases.

It may just be greenwashing to check “AI sovereignty” off the list.

When I read this thread, "Interstellar" immediately comes to mind.

Thanks for sharing!

It’s been a while since I’ve struggled with Xilinx tools, but I can’t imagine there aren’t any hardware limitations these days. Does this run on a Spartan 6, or do you need the latest UltraScale for it?

Or does this only run in simulation anyway?

A simple skill markdown for Claude Code was enough to use the local Wolfram Kernel.

Even the documentation search is available:

```bash

/Applications/Wolfram.app/Contents/MacOS/WolframKernel -noprompt -run '

Needs["DocumentationSearch`"];

result = SearchDocumentation["query term"];

Print[Column[Take[result, UpTo[10]]]];

Exit[]'

```

Works on my computer: RTX 3090, CUDA 12.6

Interesting project! I haven't really worked with Vulkan myself yet. Hence my question: how is the code compiled and then loaded into the cores?

Or is the entire code always compiled in the REPL and then uploaded, with only the existing data addresses being updated?

Very interesting! I'll definitely give it a try. However, the documentation link[1] isn't working at the moment (404).

[1] https://crux-ecosystem.github.io/MOL/