Comment by rawbot
1 day ago
In the age of AI chatbots having consumed all of Wikipedia, its relevance has waned. So I don't think they have the same pull as they did before.
1 day ago
In the age of AI chatbots having consumed all of Wikipedia, its relevance has waned. So I don't think they have the same pull as they did before.
In the recent ChatGPT 5 launch presentation, ChatGPT 5 answered a question about how airplane wings produce uplift incorrectly, despite the corresponding Wikipedia page providing the correct explanation and pointing out ChatGPT’s explanation as a common misconception.
AI chatbots are only capable of outputting “vibe knowledge”.
Yeah, but even so people use that nonsense, not checking if anything is correct. I suspect not enough people would notice that Wikipedia is inaccessible, sadly.
What is this corresponding Wikipedia page?
https://en.wikipedia.org/wiki/Lift_(force)#False_explanation...
https://en.m.wikipedia.org/wiki/Bernoulli%27s_principle
Under the Misconceptions header
Wikipedia is a moving target. Content today is not the content of yesterday or tomorrow. This is like saying all knowledge that humanity can gain has already been accomplished.
My personal test usage of AI is it will try to bull shit an answer even when you giving known bad questions with content that contradicts each other. Until AI can say there is no answer to bull shit questions it is not truly a viable product because the end user might not know they have a bull shit question and will accept a bull shit answer. AI at it's present state pushed to the masses is just an expensive miss-information bot.
Also, AI that is not open from bottom to top with all training and rules publicly published is just a black box. That black box is just like Volkswagen emissions scandal waiting to happen. AI provider can create rules that override the actual answer with their desired answer which is not only a fallacy. They can also be designed to financially support their own company directly or third party product and services paying them. A question about "diapers" might always push and use the products by "Procter & Gamble".
Its relevance has absolutely not waned, more relevant than ever. Models need continuous retraining to keep up to date with new information right?
Despite having consumed all of Wikipedia, it still can't accurately answer many questions so I don't think it's relevance or value has waned. AI has not got anywhere near becoming an encyclopedia and it never will whilst it can't say I don't know something (which Wikipedia can do) and filter the fact from the fiction, which Wikipedia does uses volunteers.
Doesn't AI essentially use the concept of volunteers as well with RLHF?
Good point, it's similar to some extent. Although clearly the quality of the work that the people doing RLHF on the major LLMs is rather low in comparison with those volunteering at Wikipedia.
1 reply →
Besides the fact that LLMs still make up stuff?
Yea great, make everyone even dumber by forcing them to use AI slop