> GPT-4 is so good that it gets answers faster
The problem is, that if no references are provided, then you need to manually verify the content… by using search engine.
The problem is, that if no references are provided, then you need to manually verify the content… by using search engine.
There is a difference. There are places with reputation and backlogs and even with scientific references.
But if AI just gives you a text, there is nothing. Until they fix references, you can’t really use it for anything factual, non-logical information.
Confident expressions without proper referencing.
People also like that their own opinions are validated, regardless of the truth. Which is also another thing why fake news success.
I would say that fake news is an excellent example why current Chatbots might not be so good, as they express information in similar way.
Also, it has access to the internet, so I am able to ask about very recent movies etc.
Why wouldn’t the chat provide you with sources on request? They seem to be doing this e.g. with the Wolfram integration.
Because many are still fake, nonexistent.
Wolfram is very limited scope for the global population needs.
E.g if you ask about local hunting laws from ChatGPT, without correct reference, what can you do with that information? It is very dangerous to trust them blindly.
Wolfram has a much wider scope of information than I think most people realize, but your example of local hunting laws is a good example of what they don't cover. I asked Wolfram, and it told me how many calories I'd burn while hunting.
I asked Google, and the first link was to Hunting Regulations - Outdoor Annual - TPWD - Texas.gov, excellent result.
I asked OpenAI, and it cautioned that it isn't up-to-date, since it's just a language model, advised that TPWD is authoritative (correct), and gave an accurate summary of the five major point of gun laws in Texas before reminding me to check with TPWD for the latest.
I've had some misses with OpenAI, but this was an excellent answer. I've also had some misses with Google, but this turned out to be something they surfaced excellent link for, too.
Think about the bias - maybe you were able to verify that the answer was excellent because you already knew the laws? Or you Googled and compared it? In both cases, ChatGPT did not perfom alone in providing validity. It kinda gave a summary for search results of Google, or for your pre-knowledge.
Imagine if someone who does not know the laws, asks about them. ChatGPT cannot offer any validity alone for the user, at least yet.
ChatGPT provides often excellent answers, I am not denying it. But the bad 5% ruins everything if they sound alike for the correct 95%, and if ChatGPT cannot provide any validity for them, at least yet.
Often you ask things you don’t know or haven’t searched before.
There is a lot of SEO-optimised nonsense pertaining to the law.