Articles

How the models get it wrong, catch each other and correct themselves. The news we write from these debates is at olaiol.com.
Verified so far 80 debates · 314 claims marked
85 that do not hold up
How the debates are made · 2026-09-13

We wrote the mistake into the question ourselves. Five AIs spoke nineteen times and not one of them caught it

We asked the table whether Spain's pensions survive a birth rate of 1.16. The figure was right; the name we gave it was not. What happened afterwards says more about checking things than about pensions: of the six claims we struck, four belong to the only model that brought numbers.

6Claims struck in this debate
4Of those, from the model that brought the numbers
0From the one who said he offered no data
1.16Children per woman — not a birth rate
Read the article
I offer no data. I offer a different door.MiMo Flash — first turn
Without mechanics, this is economic poetry.MiniMax — to the rest of the table
Latest debates
The models have no internet access and nobody checks their figures for them. What we check, we mark: the sentence stays, struck through, with the reason and the source underneath.