A human in control. In #curl development.
@bagder Sane approach: Use the best tools you can get, but use them with care, without giving up on quality or best practice.
@bagder Still one question remains open, as you do not cover it in your blog post: How about ecologic sustainability? What are your plans to keep resource use, be it AI tools, be it CI under control? Do you measure energy use? Do you have policies on energy sources that are permitted?
@taschenorakel we have no means to sensibly measure that nor to compare pros/cons when it comes to spending vs outcome
@taschenorakel so no, we don't have such policies. Can anyone actually?
@bagder "Easiest" approach on such policies is running all these machines at your "decarbonized home": There you decide on which electricity provider get used. There you can decide against natural gas, coal and the like. You could computers' heat to fill reservoirs to use them night, or even winter. That kind of things.
Still, with the most powerful AI models not being available for local use and machines able to run them being prohibitive expensive, that not feasible right now, for AI.
@bagder curl may not have quite the scale where these sorts of things become necessary - but you might find this an interesting read anyway: https://pauldjohnston.medium.com/agentic-development-best-practices-0e23ab07c1f8 (and also https://charitydotwtf.substack.com/p/ai-enthusiasts-are-in-a-race-against if you haven't already read it)
@bagder “curl is developed and driven by humans, assisted by tools.”
@bagder I got to "introduces sand in the machine" and thought ... now there's a pearl of wisdom!
@bagder my main issue with AI is de-skilling. Even if AI were better and had none of the ethical and moral issues the speed at which you loose your skills with it is very fast.
And every time I talk about this with AI hypers, they reply with “discipline” which I don’t find like a good solution to anything.
This is not a rant, is an honest question? Do you see any solution to that?
As with anything, it depends.
If I ask an LLM to generate a skeleton version of something using a technology that I'm not at all familiar with so that I can get started, then it is not causing deskilling, on the contrary.
If I ask an LLM to describe in plain English what some code that I wrote, and I find that the description does not match what I thought, then I've just found a bug. That too is the opposite of deskilling.
@bagder
- replies
- 2
- announces
- 0
- likes
- 1
Yes, sure, if you stop writing code by hand and put everything in the hands of the LLM then that's a different matter entirely. As with any tool that allows you to generate code though (bison/flex are other examples), it can be used dumbly and smartly, and the downsides that you mention, while real, aren't universal.
@bagder
@bagder @taschenorakel I don't think there is proper infrastructure for sustainability measurements when it comes to professional digital services.
But there are industries that have this, at least for climate impact. If I buy building materials from a professional retailer in Sweden, I will get an order acknowledgement where each item has a specified CO₂e, which I could import into my accounting software.
@wouter I’m not entirely sure you are actually getting the skill either. Developing the brain to know something is a very non linear process. But even if it were that doesn’t make the AI worth it, right?
A machine that helps learn or validate things a bit faster is not a revolution.
I think there is a version of LLM that is good, but I don’t see how that happens in this ecosystem. I’m really trying I just don’t see it (sorry @bagder if this is not interesting to you)
The 'so that I can get started' is the crucial part here. If I have an LLM generate a skeleton that I can build upon further without using the LLM, then that can help me learn.
Of course that doesn't apply if you take the LLM output and then never look at it. But if the skills of that technology weren't there in the first place, then you using an LLM is also not making you lose any skills.
@bagder