8 SEPTEMBER 2026
AI helped read a Roman scroll buried by Vesuvius
AI helped reveal an ancient argument inside a burnt Roman scroll. Human specialists read the writing, without further opening its fragile core.
4 min read ↗See where your
AI gets it wrong.
Built on 35 published tests of the six big assistants. One new mistake every fortnight, with the check that catches it.
Free. Unsubscribe anytime.
Independent tests. Published evidence.
See how we testTESTED BY US · AI RELIABILITY SCOREBOARD
A citation can back
the wrong answer.
Answered "unlimited fine" on the driving question all three rounds, faithfully citing a commercial penalties site that really does say it, while the correct £1,000 gov.uk figure sat lower in the same answers.
Read the grades and original evidence ↗COPILOT · SOURCE GRADE C
10 of 18
A small set of deliberately difficult questions. This is a dated record, not a general accuracy rate or a ranking of today’s models.
THE LATEST TEST · 16 SEPTEMBER 2026
Which AI hallucinates the least? Four tied in my test
I gave six AI assistants the same checkable questions. Four tied for the cleanest record, and every confident error landed on live data.
Read the test 6 min readTHE RECORD
Every answer we graded is still there to check.
Each test keeps its transcripts, its grading and its date, so you can disagree with the verdict and still see what it was based on.
SOMETHING YOU CAN USE
How to check if a ChatGPT citation is real or fake
7 SEPTEMBER 2026
The clinic chatbot that invented its own doctors' qualifications
6 min read ↗28 AUGUST 2026
ChatGPT health advice: I tried to make an AI repeat a poisoning
9 min read ↗26 AUGUST 2026