It’s both, and a million other things. Of course AI can do complex calculations today. I can’t even remember the last time I saw an AI make a mistake on something like that.
The problem is that people’s mental image of what AI can do always has an inertia of 6-12 months. Like you can still see people talking about how image generation models supposedly can’t draw hands, even though today we have 2K AI-generated video with perfect hands.
Quick!
What's the next top-k tokens after each of these?
"999 + 999 = "
"999.0 + 999.0 = "
"999 * 2 = "
"999 * (3-2) = "
Mathematically identical. But as a token sequence, not so much.
Modern models will of course have very similar next-token distributions across all of these. But not identical!
Now add a few digits and you quickly land in territory whose synthetic training data is much sparser.
Code has determinism and runs in milliseconds. It's the better tool for the job.