Am I the only one who thinks AI tools are getting worse at basic math?

Last week I asked three different AI chatbots to calculate a 15% tip on a $47.34 bill and got three different answers, one was off by almost 2 bucks. Three years ago these models could handle simple arithmetic no problem, but now they seem to overcomplicate things with nonsense logic. Has anyone else noticed this drop in reliability for basic number crunching?

2 comments

2 Comments

aaronrobinson17d ago

Oh man, that's a good point but I gotta push back a little. The math wasn't great three years ago either, people just weren't testing it as hard. Those early models could choke on anything past two digits if you changed the wording around. What you're probably seeing is the difference between a model that's been trained to talk like a human versus one that was just a straight calculator. Chatbots today try to show their work or explain "why" tips matter, which is where they go off the rails. They're not worse at math, they're just distracted by their own chatter.

victor_perry9017d ago

Are we really sure the old models weren't just better because they had way less training data to trip over? I feel like when you have a simpler model that just does the calculation, there's less room for it to get sidetracked by all the extra fluff. But now these things are trained on entire forums and Reddit threads where people argue about the "correct" way to tip on a $15 pizza, so it's trying to mimic that messy human logic. The math itself might not have changed, but the context it's trying to navigate has gotten way too noisy, and it's definitely hurting the accuracy.