There is no need to watch this chess video for the purposes of the post, or even understand chess. Suren who made the video, asked AI system ChatGPT to find him a nice chess game involving a queen sacrifice. It offered him an old game between Spassky and Bronstein which did contain a queen sacrifice.
However, ChatGPT made two factual errors. Firstly it gave the year of the game as 1960 when it was 1956. Secondly it reported that Bronstein won the game when Spassky won it. Suren knows his chess, makes excellent chess videos and is bound to be right, but to be sure I checked via another online source.
I've come across factual errors made by this system before and it could be said that it's early days so we should accept a few errors however odd. Yet a chap is bound to wonder how factual errors are corrected and who corrects them. Wikipedia springs to mind. Meanwhile -
ChatGPT: Microsoft to invest billions in chatbot maker OpenAI
Microsoft has announced a multi-year, multibillion dollar investment in artificial intelligence (AI) as it extends its partnership with OpenAI.
OpenAI is the creator of popular image generation tool Dall-E and the chatbot ChatGPT.
In 2019 Microsoft invested $1bn (£808m) in the company, founded by Elon Musk and tech investor Sam Altman.
8 comments:
We don't seem to be getting anywhere nearer developing consistent and reliable sources of truth. I used to think I knew things because I read a lot and had a good memory. I prided myself on being better than people who were less educated. But then along comes the internet, and it turns out a lot of the books were wrong, or biased, and we could look things up on the web. But that turned out to be written by the same people who wrote the books, and was so riddled with inaccuracies and churnalism and so full of noise that everyone considers themselves an expert, but they all disagree. And now it seems that the AI applications used to search the web are just as bloody useless...
In earlier days it was reported that Wikipedia did contain errors - but no worse than Encyclopaedia Britannica. While I suspect that some entries in Wikipedia have been corrupted by the Woke World View who is to say that an updated Encyclopaedia Britannica would not also be affected?
In any event I have stopped supporting Wikipedia, and will unsubscribe from any web service that succumbs to the WWV.
Sam - yes, even good nonfiction books are not easy to find because of human failings and bias. We already see the human limitations of AI which must be a related problem. Maybe the internet is showing us what was always there and what we always had to do - compare sources and be particular about what we read.
I've noticed that YouTube videos made by keen amateurs can become less reliable as those amateurs become more professional and need to attract large numbers of viewers.
DJ - I became aware of the Wikipedia editing problem when I stumbled across an entry for Galileo claiming he had over 100 children. Soon removed of course, but the editors are human with human failings and powerful organisations are always there exerting their pressure. Some demented individuals too.
It's Microsoft: you wouldn't expect accuracy, would you?
AI just churns out a melange of huge amounts of stuff it finds on the internet, as far as I can tell.
So problems like this are just another case of GIGO.
Plus ca change, etc.
dearieme - I wouldn't expect accuracy. I'd expect answers to require regular updates forever.
Peter - it seems like that to me too. It can't find what isn't there already.
In the early days of the web (say c1995) most websites were unofficial, created by techies in unregarded corners of the organisation. The presentation may have been elementary but the content was often useful, real information. Then it al became corporate and marketing etc. took over. So instead of precise useful technical documents most website now contain vague waffle.
I am not sure how much of Google's original Page-Rank algorithm is still operative but it would inevitably contribute to the dumbing down of searches, more usedless 'content' linked to similar, it crowds out the rare fragments of useful information.
djc - I seem to have noticed a decline in the relevance of search engine results over the past few years. As you say, the useful stuff is crowded out. I suspect some of the crowding out reflects popularity, but only some of it.
Post a Comment