ICYMI Late Night Linux this week had a stand-out debate around LLMs and AI in FOSS and development.
-
ICYMI Late Night Linux this week had a stand-out debate around LLMs and AI in FOSS and development. Loved that it wasn’t just a pile on and really argued out pragmatism vs purity/determinism and reliance on a supplier through both a long and short term view.
-
ICYMI Late Night Linux this week had a stand-out debate around LLMs and AI in FOSS and development. Loved that it wasn’t just a pile on and really argued out pragmatism vs purity/determinism and reliance on a supplier through both a long and short term view.
Great points all round, tho I really wanna thank @felimwhiteley for the rush of heady clarity when he suggested it’s ultimately a language preference.
Sure, there are different models with differing resource needs, capabilities, privacy levels, etc. but in the end this is about how you tell a machine what you need it to do.
When you need a machine to do something important, is it better to communicate this need in a formal, elegant, well-defined language designed by a bunch of geniuses? Or entrust it all to the vague, cobbled-together nonsense and historical hot-mess of English/$your_native_language?
-
R relay@relay.infosec.exchange shared this topic
-
Great points all round, tho I really wanna thank @felimwhiteley for the rush of heady clarity when he suggested it’s ultimately a language preference.
Sure, there are different models with differing resource needs, capabilities, privacy levels, etc. but in the end this is about how you tell a machine what you need it to do.
When you need a machine to do something important, is it better to communicate this need in a formal, elegant, well-defined language designed by a bunch of geniuses? Or entrust it all to the vague, cobbled-together nonsense and historical hot-mess of English/$your_native_language?
@latenightlinux @felimwhiteley
I’m gonna say everyone feels most comfortable in their native language, and we humans have definitely managed to get stuff done this way.
But putting aside models, integration and technical stuff, simply how much data/money/lives are we willing to bet on our ability to define a problem clearly, accurately, completely and unambiguously in our native language?
And when the stakes are above negligible - how many cycles of “oh no, shit, what I meant to say was…” and “ffs of course I didn’t want you to…” before actually writing the code is just easier/safer/better all around?
