Wikipedia founder Jimmy Wales told a packed auditorium of tech enthusiasts that ChatGPT makes too many mistakes to be useful. Speaking during the opening night of Web Summit in Portugal, Wales said large language model AI was up to 30 years from being able to compete with humans.
OpenAI launched ChatGPT almost a year ago, and in that time the chatbot has sparked a revolution in the tech sector. Companies like Microsoft and Google have gone all-in on AI and governments are exploring ways to both regulate and utilise its capabilities.
Wales is less enthusiastic about the possibilities presented by large language models — at least in the short term. He told the crowd “It is not good enough for a great many purposes," adding that it is fun to play with but when you get into using, it's “pretty bad”.
Some positive potentials
Despite his reservations and negative experiences, including asking ChatGPT who his wife married and finding it suggested multiple “plausible but inaccurate” suggestions, he did have praise for the way OpenAI signposts the likelihood of mistakes and false information.
He also wasn’t completely put off by the prospect of using AI on Wikipedia. While Wales doesn’t believe AI should write Wikipedia articles, he did suggest there were several “low-hanging fruits” where it could be helpful. For example, the not-for-profit organization recently worked with Meta on a set of AI tools that can search Wikipedia and improve citations of facts.
“For certain types of things like reading all of Reddit or all of Twitter it’s great for teaching it how people communicate, but that isn’t great for facts,” he said. When asked if he was referring to Elon Musk’s new Grok AI — trained on X posts — he claimed never to have heard of it.
Humans are better than AI
The problem with AI, he says, is that the stuff it is good at and gets correct includes content already widely written about on Wikipedia and by humans.
“If you ask ChatGPT questions about someone famous or a historical topic it is much better but that isn’t where Wikipedia needs help,” he explained. “We have tonnes of people writing about the Battle of the Bulge or World War 2, what we need is help on more obscure topics, and on those, it makes a lot of mistakes.”
Wales says for now AI is best for finding the gaps in the Wikipedia content library rather than filling them. Using AI to search across all of Wikipedia to find articles not currently well formed, or topics mentioned in other articles but not currently featured.
He said it could also be used to look for unsourced negative statements about individuals. “There is going to be some really interesting low-hanging fruit use cases,” Wales explained, but the true use cases for AI where it matches human cognition are 20 or 30 years away.
This may come as a surprise to OpenAI founder Sam Altman who claims his company is already training an AI model that could be a super intelligence and may be available in the next couple of years.
For now, Jimmy Wales is content with the thousands of “human intelligences” powering Wikipedia every day. “If you add up the computational power of brains writing Wikipedia, it far outstrips what any company can do now with large language models, he said.”