Rodrigo Arias Mallo, maintainer of the Dillo web browser, has written a blog post with a proposal on one way to ensure that a contribution is written by a human and not AI; he suggests asking new contributors to record their programming session using asciinema.

In the same way that LLMs generate patches, they can also generate the asciinema recordings themselves. Then, the contributors can lie to the reviewers pretending to have made the edits. Perhaps surprisingly, this is not a easy task for LLMs, at least from my observations. The corpus of recordings of developers making mistakes and thinking the whole process of editing a file is not as large as the corpus of FOSS programs and patches in which to train an LLM. During my very simple tests I haven't been able to generate an asciinema session that remotely resembles what I would expect from a human, and even less so from a human with a nice editor theme and editing an existing Dillo source file.

The Dillo project is not yet requiring asciinema recordings, but he said that he would like to test the theory further. LWN covered asciinema in January 2026.

Update: the New York Times has published an obituary of Dr. Neuman.

This is an interesting challenge from the “why not?” files — [GPUSpecs] over on YouTube built a gaming PC without using a single component from NVIDIA, Intel, or AMD. That immediately makes us think of the high-power ARM workstations or perhaps even perhaps the new “AI workstations” coming available with RISC V architecture, but the challenge here was specifically “gaming PC,” not workstation. A gaming PC, without a GPU by one of those three? To make it even more interesting, the x86 CPU isn’t Intel or AMD either.

If you’re of a certain vintage, you may remember Cyrix. Cyrix reverse-engineered the x86 ISA and made their own compatible chips in the 90s, before being bought out by National Semiconductor, and then VIA Technologies. VIA partnered with the Government of Shanghai to found Zhaoxin, and it is from Zhaoxin that the KaiXian KX 7000 CPU hails — an x86-64 device, that isn’t Intel or AMD. We’ve actually covered the company before. This particular chip benchmarks like an old i5, so not spectacular, but usable. 

The GPU is also Chinese: a Moore Threads MTT S80, with 16 GB of DDR6 vRAM, 4096 shading units, 256 texture mapping units, and 256 ROPs. On paper, that looks like a very respectable graphics card, but it’s not clear how well the games [GPUSpecs] tested were actually using it. Based on the numbers he was getting in his testing, there are some serious driver issues with this card. Even Black Myth: Wukong, which is supposed to be a game the card targets, was sitting at 13.6 FPS on low settings and 1080p. That almost feels like integrated graphics numbers, not something a beefy GPU would give you — but it matches what other reviewers were saying when the card first came out.

So if you’re looking for a sanction-proof gaming rig, we’re sorry to say it’s not quite ready for triple-A. On the other hand, it’s a neat hack and we didn’t know this box could even get built. Right now, it looks like you will need at least one of the big three names to game on–you can game on ARM with NVIDIA graphics,  or even with Intel graphics, and of course AMD, which has been in the works the longest.

Jon Seager, VP engineering for Canonical, has posted an update on "what Canonical and Ubuntu will do (or not) to incorporate AI" that explains what part AI will play in the future of the company and its distribution.

The bottom line is that Canonical is ramping up its use of AI tools in a focused and principled manner that favours open weight models with license terms that feel most compatible with our values, combined with open source harnesses. AI features will be landing in Ubuntu throughout the next year as we feel that they're of sufficient maturity and quality, with a bias toward local inference by default.

AI features in Ubuntu features will come in two forms: first as a means of enhancing existing OS functionality with AI models in the background, and latterly in the form of "AI native" features and workflows for those who want them.

This year Canonical has begun a more deliberate push toward education and developing competence with AI tools. We are not setting shallow metrics on token usage, or percentages of code written with AI, but rather incentivising engineers to experiment and understand where AI tools add value. Rather than force a single early-choice AI stack, we're incentivising teams to each pick 'something different' and go deep, so we learn more as an org in the next six months.