8
votes
What trustworthy resources are you using for AI/LLMs/ML education?
Every company is trying to shoehorn AI into every product, and many online materials provide a general snake oil vibe, making it increasingly difficult to parse. So far, my primary sources have been GitHub, Medium, and some YouTube.
My goal is to better understand the underlying technology so that I can manipulate it better, train models, and use it most effectively. This goes beyond just experimenting with prompts and trying to overcome guardrails. It includes running local, like Ollama on my M1 Max, which I'm not opposed to.
3 Blue 1 Brown has an excellent playlist of how AIs work. I don't know if it will help you with the practicalities of training your own models, but it is an excellent way to understand what they are, how they work, and what the training process is supposed to do.
Jay Allamar has written some excellent pieces on the serious technical fundamentals, which go deep but are still accessible to a non-expert reader with a decent technical background. I'd start with The Illustrated Transformer (or maybe his article on attention, one step before that), which I think hits the most important topics that are relevant to models you'll see day-to-day.
Two minute papers on YouTube often covers newly released research on AI/ML topics, explained in brief by an academic with a strong background in mathematics and GPU computing.
Papers With Code is where I go when I need to use a model for a specific task and don't know what's current in that particular area. Their benchmark tables aren't perfect - they depend on what tests a given author chose to do, so it can be misleading either because of cherry-picked results or because the actual state of the art model didn't happen to be tested on the dataset for the chart you're looking at - but I've still found it a very good starting point for what I should be looking at when I need to get something done.
Fireship (also YouTube) is a bit more divisive in tone - very cynical and meme heavy - but he undeniably does a good job of hitting the salient points quickly. It's a general programming news and info channel, but hits on AI topics fairly regularly because they're a big part of the tech world nowadays.
Hugging Face writes some pretty good blog posts, although they're definitely also intended for promotion, and influenced by their position in the industry. On the plus side, that position is broadly similar to GitHub's, so their interests tend to be more on technical progress and less on snake oil, even if they want to spin things in a way that keeps them as the hub of that progress.
I'm also still getting Nofil Khan's newsletter, although I'll be honest and say it normally goes into the "I'll read that later" bucket nowadays and then gets forgotten about. He was one of the earlier voices keeping on top of the rapidly evolving space a few years back and I really appreciated it then, but now I find there's a bit less need for what he does as the dust has settled somewhat and the mainstream tech zeitgeist is more on top of developments in the field.
Finally, if you want to go behind the curtain a bit, take a look at the models people are using and discussing on Kaggle. There are big financial incentives there, and the results depend on hard mathematical results with nowhere to hide any marketing fluff, so the discussion threads and sample code tend to hit on the actual state of the art models and techniques for a given task pretty quickly - and the people competing have often meaningfully advanced those models too by the time a competition is done and the results are published.
For news, I've enjoyed https://youtube.com/@aiexplained-official. He's to the point, not sensational, and makes a point of reading papers and running tests before commenting on new developments. Useful for staying abreast of new models.
Intelligence Illusion for a more general, realistic analysis of generative ml models usage in business context.
These two are good for general technical understanding :
https://www.youtube.com/@YannicKilcher/videos
https://www.youtube.com/@AndrejKarpathy/videos