pull down to refresh

Why AI is so bad at sports information (?)
808 sats \ 10 comments \ @SimpleStacker 7h Stacker_Sports
As a joke, I once asked ChatGPT to give me a 2025 power ranking for all 32 NFL teams (#1206256). It turned out that the Power Ranking was absolute trash, getting key facts wrong and even listing the same team twice and forgetting about a team!
Today, I asked ChatGPT how long Kawhi Leonard, Paul George, and James Harden played together on the Clippers, and somehow it told me that James Harden signed with the Suns after the '23-'24 season, which I'm sure is wrong!
That got me wondering, why is AI sooo bad with sports information? I understand mixing up power rankings, keeping track of 30+ teams is hard! But a simple fact like what team Harden's on?
So, I asked ChatGPT itself why it got this wrong, and it confessed that it gets confused by the massive amount of fan speculation and internet rumors surrounding sports moves.
I know that ChatGPT regularly consults Reddit for certain topics, and Reddit is (sometimes) reliable, but for sports it's almost certainly gonna get a massive amount of speculation, rumor, and even straight shitposting.
Just food for thought that some people might've been interested in. Crosspost ~AI
write
preview
related posts
72 sats \ 3 replies \ @grayruby 6h
That's really interesting. They can't decipher between legitimate facts and shitposting.
reply
40 sats \ 1 reply \ @SimpleStacker OP 6h
Yeah to me this is interesting insight into why AI is good at certain stuff and not at other stuff.
It'd good at code because for the most part no one is shitposting fake code. Most of the code that exists in public repos online are for real projects.
It might be good at some arcane academic subjects that get relatively little discussion outside of the experts, because most of the stuff posted online is by said experts.
But when it comes to something like sports which is massively popular among everyone, laypeople and experts included, and in which many people deliberately don't take too seriously (a lot of joking, fake rumors and speculation just for fun), it has a lot of trouble sorting fact from fiction.
I wonder if it will learn to approach online sources differently for different topics...
reply
30 sats \ 0 replies \ @grayruby 6h
I am sure it will eventually learn to decipher which data to weigh as more relevant.
reply
40 sats \ 0 replies \ @Undisciplined 6h
Just like my dad!
reply
30 sats \ 1 reply \ @nout 4h
Did you try different AI, like Gemini?
reply
0 sats \ 0 replies \ @SimpleStacker OP 3h
No, I didn't
reply
30 sats \ 2 replies \ @HardMoney 6h
Ai companies are building verifiable finance data apis like perplexity finance or a range of other competitors. I’m sure something similar will occur for verifiable sports data
reply
0 sats \ 1 reply \ @SimpleStacker OP 6h
But then that doesn't feel like artificial intelligence anymore if they have to plug into specific APIs. It's just a natural language interface around a more traditional data service.
@optimism
reply
0 sats \ 0 replies \ @HardMoney 5h
Like this but a sports statistical record search
reply
0 sats \ 0 replies \ @SimpleStacker OP 5h
Also, sorry ~Stacker_Sports, this is SN -- you can't avoid nerdification even in your jock territory!
reply