> In tabulating the “errors” I saw the most astounding result I have ever seen from an LLM, one that made the hair stand up on the back of my neck. Reading through the text, I saw that Gemini had transcribed a line as “To 1 loff Sugar 14 lb 5 oz @ 1/4 0 19 1”. If you look at the actual document, you’ll see that what is actually written on that line is the following: “To 1 loff Sugar 145 @ 1/4 0 19 1”. For those unaware, in the 18th century sugar was sold in a hardened, conical form and Mr. Slitt was a storekeeper buying sugar in bulk to sell. At first glance, this appears to be a hallucinatory error: the model was told to transcribe the text exactly as written but it inserted 14 lb 5 oz which is not in the document.

I read the whole reasoning of the blog author after that, but I still gotta know - how can we tell that this was not a hallucination and/or error? There's a 1/3 chance of an error being correct (either 1 lb 45, 14 lb 5 or 145 lb), so why is the author so sure that this was deliberate?

I feel a good way to test this would be to create an almost identical ledger entry, but in a way so that the correct answer after reasoning (the way the author thinks the model reasoned) has completely different digits.

This way there'd be more confidence that the model itself reasoned and did not make an error.