Fuck AI

One of the best approaches I saw was a teacher who assigned students to have ChatGPT generate a paper on a topic, and then write their own paper critiquing ChatGPT's paper and identifying the errors that it made.

What I meant with my comment is that AI is a far broader field than just LLMs. But I see so many proposals that are just a horrible waste of ressources.

For example, image analysis. A friend of mine helped to develop special tools for glacier analysis via satelite images. They trained a model to specifically analyze satellite images and track the "health" of a glacier in near real time.

Or take mathematical analysis. Some suggest to just throw a pile of data into a LLM and let ChatGPT make sense of it. But a far more reasonable approach would be to learn about different statistical models, learn how to use the tools (e.g. python), and build a verifyable, explainable solution.

I work in networking and InfoSec, and all the vendors try to add AI chatbots into their firewalls and wifi access points. But many of the challenges are just anomaly detection, or matching series of events to known bad events. But guess what all these AI tools are not: LLMs. (Except maybe for spam filters, thats where an LLM might be a good fit. But we don't need a huge, expensive model for this).