@informapirata @aitech D'accordo sullo spirito generale dell'articolo, ma poi mi perdo nel dettaglio. C'è una ragione tecnica per cui i LLM sono definiti come inaffidabili per scrivere, ma affidabili per correggere? Se, come dice l'autore, fa errori quando gli fai scrivere un paragrafo in autonomia, perché dovrebbe essere più efficace nel segnalarti gli errori che fai tu? Intendo proprio errori fattuali e interpretativi (quelli per cui ti serve un editor vero), non grammatica e refusi.

"If you didn't take the time to write it. Don't expect me to take my time to read it."

LLMs are destroying our cognition.

I'm noticing my own company doing this now. Everything that gets shown to potential customers is so obviously AI generated, and it all looks cheap as fuck.

I don't know how the leaders in the company fail to see how sloppy everything has become... Everything looks like slop. It's beginning to be embarrassing to work here. Yet... Everywhere looks similarly bad.

The tech industry is so fucked right now. It's going to collapse on itself even if the US government considers ai too big to fail and props them up. No one wants to buy slop. Why would they?

A coworker sent me an eight page summary which could be boiled down to simply “change part number A for part number B”. I wrestled with it for a while thinking “How unusual for someone to write a report for just me to read” before I realized I was looking at AI vomit.

We were talking over voice chat at the time and I had to scramble to obfuscate the sheer contempt in my voice after saying “Did you just send me an AI report?!” and keep things civil.

I later complained to another coworker about it, in vague terms without naming names because I don’t need office drama, and they immediately knew who I was talking about because there’s only one AI fan in our group. Whoops.

Anyway I’m not reading an eight page report you probably didn’t even read before sending it to me.

After writing a paragraph, many times I would ask the AI to ensure what I had written was correct by referencing the source code or configuration, or examining production logs or metrics. For example, ensuring I had correctly described which columns in the database were indexed, or how the rows in the Parquet files were sorted. This allowed me to stay in the context of writing, often moving on to the next paragraph while the AI verified the previous one. It helped me spot important omissions and inaccuracies.

"Am I a joke to you?" - Editors, probably

Even this is LLM generated? It is not—it is!!!