Matrix Games Forums

Ripley: "Hudson, this ChatGPT bot has retrevied and collated info of the last twelve years from this obscure game forum!"
Hudson "Why don't you put *it* in charge of Jap AI?"

My impression/opinion of the recent trend in AI is that a lot of people don't understand what ChatGPT actually is and what it can and cannot do.
A lot of people seem to trust the results a bit too much, ChatGPT can often very confidently present you information, that is completely wrong.(They got better, but it is still there and the issue will not be resolved anytime soon.)
Results from that bot should always be verified, but that is something often not done.

To put it another way: Wikipedia can be incorrect, inaccurate or ambiguous at times, when it comes to historic articles. ChatGPT knows the content of Wikipedia, but that is also there it's limitations are, it cannot possibly know more about historic events, than what is written in Wikipedia. So it will happily return you all the false information that is on that site and confidently present them to you as "the answer".
The problem is that this is not something that can be solved by ChatGPT because the raw data itself is false. The Quality of the answers ChatGPT can give is directly related to the Quality of the raw data. There is also no way of knowing where the raw data, for the answer you got, was collected from as the bot doesn't provide this info.
Lastly, ChatGPT is not actually intelligent. The answer it returns was not given because ChatGPT drew the logical conclusions necessary to come up with the answer. It learned a probability, through rigorous sessions of trial and error, what the answer to a question could be, given a set of input parameters.
That is why answers from bots like ChatGPT always need to be verified and not blindly trusted.

I am not aware if there exists a paper already on that matter, but it would be interesting to know what the current and past error rates of ChatGPT are. I am sure that OpenAI at least roughly knows the numbers...

So no, I honestly wouldn't say that ChatGPT is making these forums obsolete. First, without this forum ChatGPT wouldn't even be able to answer any question about WITPAE. Secondly I don't think the Matrix forums are a mere platform to ask questions and get an answer for them. People here socialize, tell their in-game or private stories, happily join discussion about various topics(like this thread), find opponents for matches or just take pleasure in writing advice to other players. That is not something ChatGPT is able to replace.

All in all, it is indeed impressive what OpenAI was able to create with ChatGPT, this is not something that existed previously. But in my opinion, people are little bit too hyped about it and are trusting it results far too blindly. I am not even addressing the other issues this and similar tools have, like the huge ecological footprint, the bias in the answers often expressed as racism, lacking security standards, the start of widespread use of ChatGPT through Scammers, the possible start of repression of original creations through mass quantity AI creations,...
Currently these new AI technology seem to be creating more problems than they solve and I am afraid the answer to all these problems will be just another AI, with a whole set of new problems.

Well said.