Ancestors

Written by Feoh on 2025-01-23 at 15:19

An excellent example of where LLMs can save time and introduce near zero risk:

"Convert these Python data structures to YAML".

(I had plopped a couple of pages of Python syntax into a YAML template and didn't want to navigate the reformatting manually, since I find YAML excruciating to write. Yes, I know I'm strange in this regard and everyone loves it. I enjoy JSON too. Break out the tar and feathers :)

=> More informations about this toot | More toots from feoh@oldbytes.space

Written by S. Huff on 2025-01-23 at 17:47

@feoh i think i’m missing some crucial context here, since my answer to that same question is:

https://yaml.dev/doc/ruamel.yaml/basicuse/

what am i missing? where is the value-add from the LLM? i mean that sincerely.

=> More informations about this toot | More toots from hakamadare@wrong.tools

Toot

Written by Feoh on 2025-01-24 at 01:43

@hakamadare this may not fly for you, but the value out is quite simply convenience.

sure, I can install a YAML Parts are into whichever Python I happen to have it hand, write the code, process the file, and use the results.

Or I can simply copy and paste the blob of code I want converted into a different format and have the LM do it with no project, no code, and no external dependencies installed.

heaters are gonna hate and so I don’t pretend that this is going to convince you the AI is useful if you’re in the opposite camp, but for me, this is a fine example of where I can save time and key strokes with very little risk if any.

=> More informations about this toot | More toots from feoh@oldbytes.space

Descendants

Written by mneme on 2025-01-24 at 04:46

@feoh Sure, but it's a case where you could use the tools and do it easily with no risk, or use a LLM and do it with only a little risk. Given the minor upside I'm not seeing it being worth the risk.

=> More informations about this toot | More toots from mneme@dice.camp

Written by Feoh on 2025-01-24 at 14:38

@mneme You're generalizing, and that's unfortunate because it makes your argument not stand up to this particular set of circumstances.

I know Python really well. I'd class myself as a journeyman Pyhonista with aspirations of becoming an expert.

YAML is a format which, despite its vast scope and numerous frailties is very easily understood in the simple case which this was.

So there is provably exactly zero risk in pasting a blob of configuration data (not even actual algorithmic code in this case!) into an LLM and asking it to spit out YAML.

The benefit I get is time savings, as I previously stated, were I to use your method I'd need to:

=> More informations about this toot | More toots from feoh@oldbytes.space

Written by S. Huff on 2025-01-24 at 14:14

@feoh no shade intended!

for me this course of action could provide meaningful convenience if the context were, say, that i had to work in Erlang; i’m not at all familiar with that language or its ecosystem, i’d have to do a lot of reading. but Python?

i guess this just reminds me of when i first got a cell phone that could save numbers: my initial reaction was “why do i need this, i’ve already memorized all of my friends’ numbers?” now i barely know anyone’s number by heart.

=> More informations about this toot | More toots from hakamadare@wrong.tools

Written by Feoh on 2025-01-24 at 14:32

@hakamadare None taken!

And yes I TOTALLY agree.

I think there's a lot of that, and I also think that many people who work in our field, whether they'd admit it or not, have a tendency to choose a suite of tools and worship them as the Holy of Holies, not to be supplanted, replaced or augmented by any other.

(And if that sounds far fetched to you, tell me you DON'T know a sizable group of people who still use Vim, code in nothing but C and consider UNIX to be the second coming and the ultimate evolution of technology for all time :)

I pride myself on being a pragmatist. I make a point of trying out new tools, and ESPECIALLY whole new types/categories of tools as they become available.

I purposefully don't TRUST llms and am very careful to constrain the ways in which I use them to areas where I am 1000% sure I can fully understand what they produce, or at least the results that their output produces, but I do find them useful and convenient. There's nothing magical about them. They're just a way to save time and effort in areas where it makes sense.

=> More informations about this toot | More toots from feoh@oldbytes.space

Written by S. Huff on 2025-01-24 at 14:49

@feoh

(And if that sounds far fetched to you, tell me you DON'T know a sizable group of people who still use Vim, code in nothing but C and consider UNIX to be the second coming and the ultimate evolution of technology for all time :)

=> View attached media

=> More informations about this toot | More toots from hakamadare@wrong.tools

Written by 『-𝚍𝚜𝚛-』 on 2025-01-24 at 15:06

@feoh @hakamadare

Here's my contributory point:

=> More informations about this toot | More toots from dashdsrdash@tilde.zone

Written by Feoh on 2025-01-24 at 15:08

@dashdsrdash @hakamadare You're absolutely right, and using an LLM to convert a non trivial Python data structure into YAML is just stupid, because you then can't easily verify the output with a quick read.

I use and like using LLMs, but I NEVER trust them, and neither should anyone.

=> More informations about this toot | More toots from feoh@oldbytes.space