Obviously.

People who produce large documents with LLMs seem to frequently be under the impression that the recipient would like to receive a large document

They are pretty much always wrong

That's what you get for being a recursive acronym, PHP!

Meas mór agam ar lánúineacha a thoilíonn an dara páiste a bheith acu. "Bhí an chéad cheann strusmhar, déanaimis arís é!"

You know your students are paying attention when they keep asking questions you were planning to answer a few slides later.

So that's where all that ugly graphics in the world is coming from!

RE: https://mastodon.social/@simple_sabotage/117430086608217362

Are you working for a company that makes something you kind of disagree with? We have some advice for you.

“I’ve been doing AI since when it was expert systems.” – Michael Iantosca https://resources.gala-global.org/ai-content-chain-production-localization/

When the people behind DocLang https://doclang.ai/ claim that each of their XML tags maps to exactly one LLM token, how can they know that? Surely each LLM, or each family of LLMs, tokenizes differently. Or what am I not understanding here?

“Don’t be evil” is an aspirational goal, “try not to be evil” a realistic one.

I wonder would people still be buying me coffee online if they knew I can get free coffee in the office where I work? https://buymeacoffee.com/michmech

“A paper about turning papers into agents has turned itself into an agent. The recursion, it seems, has already begun.” https://spectrum.ieee.org/paper2agent-ai-agents-research-papers

In the early 2000s when I was working for Microsoft in Dublin, the employee restaurant there became the location of Starbucks’ first coffee shop in Ireland.

In 2004/2005 I changed jobs for Dublin City University while, almost simultaneously, Starbucks opened their second Irish coffee shop in the campus restaurant there.

The pattern is clear, isn’t it? If you don’t want a Starbucks to open near you don’t offer me a job.

Seriously considering charging my clients per token. Reasoning tokens included, of course. And no caching!

I'd like to see some evidence that documents formatted in this new DocLang thing are actually understood by LLMs better than other minimalistic formats such as Markdown and YAML. https://www.theregister.com/ai-and-ml/2026/06/16/a-modest-proposal-reformat-everything-to-make-documents-more-palatable-to-ai/5255938

It's not a bug. It's not a feature. It's an emergent property.

When somebody asks you what your nationality is feel free to make something up because frankly they all are.

Next time somebody asks me a question that requires some thinking, I’m going to ask them to perform a short unnecessary computation first to prove they are serious. Yes, I am implementing #ALTCHA in my user interface!

RE: https://mastodon.ie/@helvick/117293749739311510

The original AI.

Unkrautbeet