On Fri, Feb 20, 2026 at 11:23:09AM -0500, Theodore Tso wrote:
But as you say, for some people the arguments they make against the use of "AI" (whatever the hell that means) is about the consequences, either real or imagined or projected, that they think the technology will have.And the problem here is that it's still early days, and some of the initial estimates about (for example) water usage of AI queries seem to have been half-baked, or at least, highly disputed. For example, instead of a 16 ounce of bottled water, there are other estimates that put it at 45 mililiters, or even 0.26 milliliters. One of the reasons behind the wide disparity is about how to account for the cost of training the LLM. That's a one-time cost, and as the number of queries go up, that cost will get amortized and go down significantly. It's also the case that if you believe that the LLM is going to be trained *anyway*, the water savings of foregoing the use of the LLM for code reviews, or cherry-picking of a patch, or even vibe-coding is probably going to be in the mililiter range as opposed to the entire bottle of water range. I think people also tend to under-estimate how much water gets used when creating their pair of jeans (estimates range from 4,000 liters to 10,000 liters), or how much water they use when taking their daily showers, or how much CO2 they produce when travelling to conferences or visiting their families. (OMG! An Open Source maintainer is travelling all over the world to go to conferences! Let's ban the use of their software in Debian!! *That* will solve the global warming crisis....)
Another interesting comparison is resources spent on CI. So far I see that both OSS and commercial development adopts more and more CI and QA tools, spending more and more CPU cycles on running them during the development. That, of course, is true for Debian too.
-- WBR, wRAR
signature.asc
Description: PGP signature
