Dumb Pajeet
(media.scored.co)
New here?
Create an account to submit posts, participate in discussions and chat with people.
You are viewing a single comment's thread. View all
Toast message
Speed is another advancement. This spreadsheet is mission-critical for our sales team, and in sales, timing matters. The faster you follow up, the better your chances of closing a deal.
In short, AI gives us more time and more revenue.
Right.
…
[*What increase.*](https://arxiv.org/abs/2507.09089) We have to **completely redefine all the metrics** by which “productivity” is [calculated](https://mindmatters.ai/2025/03/why-llms-are-not-boosting-productivity/) across dozens of fields if we’re going to consider “press a button and spend a gigawatt to create a hallucination in seconds” with respect to historic forms of human productivity. They just can’t be compared 1:1.
>One of my team members used to spend six hours building a spreadsheet every week. We trained an LLM using the before-and-after versions, and now it completes the same task in under a minute.
Sounds painful. Having no context to the type of document being created (I understand; you should protect yourself), that *still* sounds like something that could have been done for the last few decades using a custom-made form template in [FileMaker Pro](https://www.claris.com/filemaker/), such as those created by a local medical facility where I interned nearly 20 years ago… OH FOR FUCK’S SAKE, THEY PUT “AI” INTO FILEMAKER NOW. FUCK ME, THEY’VE TAKEN [CLARUS](https://en.wikipedia.org/wiki/Dogcow) OUT BACK AND SHOT HER.
Back in my day, we did it by hand but we only had to do it by hand ONCE. I’m curious now, though. Why recreate it from scratch every week if it’s so common?
>The faster you follow up, the better your chances of closing a deal.
Of course; that’s not to be discounted. It’s the *universality of application* with which I take issue. There is no “general purpose humanoid robotics” yet. As such, you can’t take a car manufacturing robot and put it on a furniture assembly line and expect it to operate correctly. Similarly, there is no “general purpose large language model.” As such, you can’t take a generic LLM trained on stolen books, scientific papers, social media posts, personal e-mails, and GitHub repositories and expect it to be able to create compelling fiction, corporate documents, efficient software, and nonfiction essays without it shitting the bed.
How do yoy know it isn't feeding nonsense data, randomly?