tangeli

Lawyers getting caught using AI and submitting errors to the court has been a big problem for years

It wouldn’t be a big problem if the unreliability of LLMs were part of ongoing professional development, use of LLMs without thorough fact checking was clearly barred by professional ethics, and lawyers were immediately disbarred for presenting false evidence if it came from an LLM.

It remains a “big” problem because it isn’t big enough for anyone to do anything about it, beyond criticizing and ridiculing lawyers who do it.

That’s surprising. I have an older version of coreutils (9.7 from Debian 13) that doesn’t produce the links and formatting. I’m running bash and the behaviour you’re experiencing isn’t reproduced.

Aliases aside, I wouldn’t expect the shell to be the cause. None the less, I installed version 4.0.2 of the fish shell and ran the same command in it and didn’t see any control characters. But that’s not really surprising - my old version of ls doesn’t produce them. So I ran ls | less because even with my version ls colour highlights files by type, but in the pipeline to less, there were no codes.

If I run ls --color=always | less then I do see codes in the less output.

You might check your environment variables to see if CLICOLOR_FORCE is set.

Otherwise, I am out of ideas.

I would say your ls shouldn’t be working that way: when it detects that it’s standard output isn’t a terminal, it shouldn’t be outputting the control codes. Which makes me wonder what version of ls you are running.

edit: what is the output from alias ls and what happens if you run \ls --help | less?

I don’t know him. What’s a Q?

It’s true, but we’re all conflicted and imperfect. Give him credit for being open to other ideas, despite the conflict.

While unusual and on the face of it perhaps not very relevant to a project to develop an operating system, in the context of freely and generously sharing the project, explicitly open to all, including all religions, and a meritocracy, I don’t find sharing one’s sincere beliefs offensive. While the focus might be on computers and software, the open source community is, first and foremost, a community of people with values that lead them to share their work. As long as he is sincere in his beliefs, I respect his promotion of them, even if I don’t entirely agree with them. Better that than the likes of Bill Gates, Larry Ellison, Elon Musk and so many others.

I agree as far as responding to questions and issues that arise from the contribution but my understanding is that from a legal, copyright perspective, if AI generated the code then the contributor can’t own the copyright, nor assign it. So my preference would be: no AI generated code. Other use of AI tools (e.g. finding vulnerabilities) might be OK but there are still the environmental and ethical issues to be concerned with, so I remain ambivalent. It will be interesting to see the decision of the Debian community.

Even if AI generated code can’t be detected reliably, it is good to have a clear policy to guide projects and contributors. At least those contributors acting in good faith and honestly will do the right thing, and projects will have a basis for rejecting contributions from people that are found to willfully or negligently violate the policy. Blocking all malicious actors will be an ongoing effort, whatever the policy.

Maybe shares should be ‘sold’ on subscription, like so many other things these days. You ‘buy’ them but you don’t ‘own’ them - you just get the benefits of having them temporarily. That should reduce the incentive to ‘own’ shares in companies that are not profitable.

Destruction of precious resources in order to create a monopoly is not good for society.

It’s best for the AI companies if there are no sources other than their AI.

I don’t mean to disparage anyone’s efforts. I linked the WikiPedia page to provide the context. I hope the interest grows and they achieve a stable release.

Yes, exactly. Laydybird was started in 2019 and only this year is expected to publish an alpha release and hoping for a beta release in 2027 and first stable release in 2028, confirming that it is too difficult. And what are the chances that those targets will not slip? And how much will the specification change in the meantime? It shouldn’t take a decade to write a conformant browser. A simpler specification would allow more projects to deliver compliant browsers.

A significant part of the problem is the W3C making the standards relevant to web browsers and servers so complicated that the purported objective of allowing anyone to build a browser is defeated. While it is technically true that anyone can build a conformant browser, it is practically impossible for anyone without the resources of companies like Google, Apple and Microsoft to keep up. New, independent implementations are almost impossible due to the technical complexity and the anti-competitive practices of the incumbents.

According to this report:

As of 2026, three engines account for virtually all browser rendering globally.

Those are: Blink, WebKit and Gecko

What’s needed is new standards that restore the practical possibility for others to develop a compatible browser. That would help resolve or avoid this problem and many other problems that are similarly rooted in vendors attempting to control the user’s data and experience.

The biggest threats I am concerned with are:

• my liability and harm to my reputation resulting from unauthorized abuse of my systems
• consumption of my limited Internet access bandwidth and capacity
• corruption or loss of my systems and data
• theft and sale or other abuse of my private data

The first results mostly from the risk of my systems being incorporated into bot nets.

The second from bot nets and abusive crawlers.

The latter results mostly from ransom ware or open ended theft and exploitation of my data.

The threats are not all from the Internet. I saw an article recently about ‘smart’ TVs that are configured to be part of a web scraping infrastructure that the vendors sell as a service, in some cases configured to use up to 200GB per month of bandwidth.