One question you may wonder these days is whether AI is completely useless and a total waste of resources, or not. After all, you can't trust what ChatGPT says, and if you can't rely on your AI chatbot being right, what is the point of having an AI chatbot? You ask the LLM something, it guesses an answer, and now you have to make sure the answer is right. The amount of work you had to doesn't seem to have decreased, only changed: now your job is making sure the LLM got the answer right.
In my opinion, AI isn't useless. AI is an alternative, and alternatives are good.
Before AI we had search engines, and before search engines we had link directories. If you can't find information using these traditional methods, you can try AI, and, if you can't find information using AI, you can try the traditional methods.
In fact, these are several alternative search engines you can try if you think Google and Bing aren't being helpful. AI chatbots are just another alternative you can make use of.
Alternatives don't need to be perfect replacements. In fact, it's better if they aren't. After all, if every website became a TikTok clone, where would I go for a feed that isn't full of short videos?
It's not a good idea to compare AI chatbots to traditional search engines. You can't use them like you would use a traditional search engine, nor can you take their output like you would a SERP. Interestingly, AI is particularly useful at the sort of query that search engines are worst at.
Just because AI is flat out wrong sometimes, and they can't be relied upon to give the correct answer, that doesn't make them completely useless. It just means that you shouldn't use AI for this sort of thing. The sort of thing that they are bad at.
Personally, I feel the problem of AI making things up, also called "hallucination," is fundamental to how LLM's work and it's never going to be "solved." In fact, in my opinion the whole paradigm is backwards. The public believes hallucinations are bad, so AI companies invest massive amounts of resources trying to fix what is an unfixable problem. Hallucinations are never going away so long as LLM's are LLM's. What I believe to be the right way forward is to give up on the pretense that you can ever rely on LLM's to be correct and embrace hallucinations as a feature of the software.
That doesn't sound like a useful feature, because we want something that gives us correct information. But if we didn't want correct information, but something else, ideas, inspiration, for example, then it doesn't matter if the information is correct or not. Hallucinations can be a fine source of inspiration just as anything else. Inspiration doesn't need to be only creative. Maybe you're inspired to do things you wouldn't otherwise, to search for information that you wouldn't otherwise.
In other words, when we think about AI as a tool that gives you accurate answers, then it is indeed useless. It's unreliable and random. You can never trust. But if you think of it as a tool that just generates random text based on a topic, it can be useful in its own way.