On 3/27/26 6:03 PM, Moses McKnight wrote:
[snip]
Here's an article I just saw. While they are not sure why things
changed about a month ago, I think from what I've seen the biggest
change was ChatGPT-5.3-codex and now 5.4, and Sonnet/Opus 4.6.
https://www.theregister.com/2026/03/26/greg_kroahhartman_ai_kernel/
Yes, LLMs are very good at statistics and pattern matching. That has
been fine-tuned to detect many subtleties in code that are otherwise
only visible by very hard looking eyes. I'm all for using this type of
machine learning to improve code quality. However, this is not what you
do or get with nearly any commercial LLM. Anyhow, as mentioned earlier,
you do not need to be an expert to see the problems in the LinuxCNC code
base (sorry, but true). But this is not what I am talking about.
It is not a discussion about what you privately should or should not do
or use or not use LLMs. It is how it plugs into the LinuxCNC project. It
is a discussion about what Github/Microsoft a) are generating code with
their system that has a questionable/unresolved copyright status and b)
about them collecting user data and putting that also in the mix. They
use and abuse any data to their liking and profit while anybody caught
up in the web of the LLM entrenchment will be left alone to solve the
problems that may (will) occur.
To exemplify *why* this is a problem should be very clear after reading
this story:
https://www.neowin.net/news/microsoft-copilot-is-now-injecting-ads-into-pull-requests-on-github-gitlab/
And this is only the beginning and it never ever ends well for the
common user.
--
Greetings Bertho
(disclaimers are disclaimed)
_______________________________________________
Emc-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-developers
