Curious find of the day: the AI community on X has seemingly re-discovered the World Wide Web, in that it now favours HTML over Markdown for “agents to communicate with us”: https://x

[joakimfors@mastodon.green]

@ricmac

[tinmouth@infosec.exchange]

@ricmac html! Who'da thought?!

[ricmac@mastodon.social]

Unfortunately, I think the next step might be: “What if we created a virtual DOM…”

[fionasboots@tech.lgbt]

@ricmac How much does handling 12000 (or 6000 for that matter) tokens cost? Do they need to submit the entire thing each time to get some sort of result: a fix, a change, extra code, reformatting, etc? What happens if you send an LLM just a tiny section of code without context? Does that work (I would assume either not at all or not very well) ... so does that mean that people using AI are burning through 1000's of tokens at time as they iterate through development? That sounds crazy expensive and wasteful!

[bms48@mastodon.social]

@ricmac @fionasboots This is the same bollock (as we say in Scotland, quaintly) all over again. The so-called "grep tax" and the cardinality of code residing within individual files contributing to filling the LLM's context window. It contributes towards "babysitting needed to get token-based generator to work" vs actual productivity. But hey! Time to make agentic automation scalable! /s

[synlogic4242@social.vivaldi.net]

@ricmac wake me when the vibe coders reinvent/rediscover the use of programming languages to, oh I don't know, specify the desired behavior of software

[glitzersachen@hachyderm.io]

@ricmac

Wow. It's such a sh*t show.

"AI" (LLM style) is really a gathering point for the clueless, who then make important self-aggrandizing noise.

[glitzersachen@hachyderm.io]

@fionasboots @ricmac

All questions that should not need to be asked. If I need to ask those questions, the technology is crap.

Can we now get back to real programming? (I don't mean you. This utterance is directed at the world at large).

[misty@digipres.club]

@ricmac Unfortunately they haven’t yet discovered the concept of a “blog” where you can store posts without paying x dot com $40 a month

[glitzersachen@hachyderm.io]

@svelmoe @ricmac

You could ask an AI to create more of them especially for this occasion.

But pay attention to your token use. Maybe you can use the saved style sheet tokens for this.

[julientaq@mastodon.social]

@ricmac be ready for everything canvas.

[mikalai@privacysafe.social]

@pikesley @ricmac
reinventing the wheel .... after firing bunch of carriage keepers.

[davatron5000@mastodon.social]

@ricmac Wait until they find out about web components

[jake4480@c.im]

@ricmac

[deadwisdom@fosstodon.org]

@ricmac This seems like the funny thing about AI token optimization is gives them a direct measure for good coding practices. I can't wait for them to be like "I found this awesome hack where I go deep and learn the tech so that I can give the AI better feedback. It saves so many tokens!"

[superflippy@mastodon.xyz]

@ricmac I feel like the old witch in the woods who knows all the secret spells who everyone ignores.

[residuum@digitalcourage.social]

@misty @ricmac
How many tokens does it cost to generate this thing called "blog" from a post on X?

[dalcacer@mas.to]

@ricmac @bkastl They came full circle.

Also, they were meant, when Gordon Ramsy performed the idiot sandwich.

[mit_scharf@social.anoxinon.de]

@ricmac What is a 12116 token HTML article? Haha, I'm so out of the loop on this topic, and it makes so little sense to me

[fionasboots@tech.lgbt]

@bms48

@ricmac I don't think grep deserves to have some perverse AI failure named after it! That aside, having found out what that term means, it just lends more credibility to the idea that humans and real intelligence are just better at this. LLMs are just for corporations to make money without producing any of their own value.

If you need an actual answer, as the person who wrote the code the LLM was trained on. Don't burn through tokens (in a way that isn't transparent) to come up with slop.