It is hard to talk about self-taught machine learning over the last several years without Krish Naik's name coming up, because his channel has become a kind of free public university for the field, and a remarkable number of people working in data today have him to thank for at least part of their grounding. What I rate most is the sheer span and the angle he takes. There are playlists on statistics, classical machine learning, deep learning, natural language processing, MLOps, and more recently a steady stream of generative AI and large language model content, and unlike a lot of academic material he tends to explain things the way a working practitioner would, with an eye on what you actually do with a technique rather than only the theory behind it. He keeps pace with where the industry is going, the community around the channel is active and helpful, and the price, which is nothing at all, makes it an almost unbeatable resource to lean on.
The honest trade-offs are exactly the ones you get with free YouTube learning. There is no syllabus holding your hand, so the burden of sequencing falls on you, and a true beginner can easily end up jumping between videos that assume different starting points and feeling lost without a clear path. The production is inconsistent, the audio can be rough on older uploads, and some explanations move quickly enough that you will be pausing and rewinding, which is fine for a motivated learner and discouraging for a nervous one. There is also no assessment, so nothing checks whether the material actually landed.
The way I would use it, and the way I would advise most people to, is as a powerful companion rather than a sole teacher. Pair it with one structured course that gives you a spine to follow, and then reach for Krish Naik whenever you want a concept explained a second way, a practical walkthrough of a tool, or a quick route into something current like MLOps or building with language models. Used like that, it is one of the best free things in the whole space.