The rise of Model Fatigue – or is it just me?

As someone curating a newsletter and dabbling in AI, I am feeling both overwhelmed and bored with news about yet another AI model being released by Company XYZ that will be a “game changer” and “leaves the others in the dust”. It feels hard to guess what I should be excited about. The size of the model? Who owns it and what it costs to use? It’s terms and conditions? What it is good for? If I can use it although I live in Europe?

If I check Cursor’s list of possible models I have no idea what each of them mean and it feels weird to see minor versions of each…

It doesn’t help that the names of the models and their descriptions on Huggingface don’t make much sense to me or anyone who isn’t deeply involved in Machine Learning. And it doesn’t help either that news outlets and company marketing blogs don’t stop covering us in hyperbole headlines about them instead of selling them through case studies.

This is nothing new. We had the same with AJAX libraries, frameworks and CSS libraries before. But if we consider the amount of energy and computation power that goes into training and weighing models this seems a lot more wasteful. What we need is fewer news about models and more information what each of them is good for. Right now, it feels much more like a size competition rather than a competition of which is more applicable. It also doesn’t help that the few benchmarks we have continue to be rigged and skewed. This is something we already had during the browser wars, so thank you, but no.

I’m much more excited reporting and learning from case studies of people who used different models and found one or the other more appropriate. So, if you have those, please don’t hold back posting these.