Many have been touting the education revolution that has supposedly arrived whereby you can use AI to learn new things faster and better than older education methods. Elon Musk said that "One of the most important uses of AI is helping people learn faster and more effectively." I think this is completely true already for cases like coding. Ask it to help you learn Haskell or how to create a Node server and it has no problem, granted that you already understand core Software Engineering concepts. It will walk you through each step and explain everything in detail if you like. Software Development has truly been revolutionised by AI.
Where it falls down compared to quick and free resources like Wikipedia is pictures, diagrams, and reliability. If I want to learn how a Transistor works, why would I ask an LLM? I would have a far easier time just googling it and getting actual and factual easy to digest information. LLMs by contrast are often walls of text with a few hallucinations thrown in. And you have to know how to prompt it correctly! Below is some examples of me trying regardless:
Grok fails, ChatGPT gives a generated image which is even worse, Gemini actually supplies real diagrams, but it’s still crap because they are shared but not being utilised to help the reader learn. By contrast Wikipedia (yes it’s biased, but in hard science is great for overviews) has an explanation with beautiful diagrams, photos and tables to clearly explain how a transistor works, what they look like, and their history. How can LLMs possibly compete with this? Not only that, people on the right have complained of Wikipedia's left leaning tendencies and sometimes outright lies but at least they try to fix errors and supply sources. AI will have a bias, lie, and it will be harder to spot or rectify it so it doesn’t happen again. In fact, we haven't found a way to stop them lying yet.
All this information is better and delivered just as fast as any LLM
But while AI firms have been working hard on reducing hallucinations since the beginning, there seems to be no movement on increasing the amount of visual aids. I was momentarily impressed when I tried NotebookLLM by Google in which you can upload many sources and then ask the AI questions; but I quickly realised that it still wasn't actually helping me learn, it's just too much text. It shouldn’t be that hard to incorporate visual aids properly because I know from experience that LLMs are perfectly capable of finding images and understanding them. I have asked ChatGPT to help me with problems before and when it responds so intelligently to the photo I attach I never fail to be amazed.
This is really underutilised in my opinion, it knows it's a transparent cap!
Instead lately most AI firms have doubled down on pure text responses with tools like Reasoning and DeepResearch/DeepSearch. Apparently you can ask it a real hard question that requires a lot of resource gathering and it can do it for you in the fraction of the time a human can. I tried it, I was underwhelmed. Ask it even basic questions that most could answer with Google and less than a minute and it will still throw obviously wrong answers. Even with all that extra computing power being used I still prefer learning about anything not related to coding from established resources.
Only this week CEO of Microsoft Satya Nadella and Elon have both announced opening games studios entirely powered by AI. If this is possible, maybe learn to walk before you can run. Have ChatBots become true teaching assistants. I do think it is possible, and then maybe you can take a stab at the holy grail by making Google.com obsolete. For now Search Engines and Wikipedia will stay the favourite for people who want to learn something quickly and efficiently.
