Thanks! It is part of the clustering step, I tell the model to make a judgement of whether something is inanimate or hallucinated (as defined by low support from only non-frontier models / judgement). I iterated on this a lot and made an eval set out of my LinkedIn contacts where I run GPT5.5 with web search and xhigh reasoning to determine pseudo ground truth. I tuned this to be higher recall (more things classified as non-hallucination) but it is definitely not perfect

In my case the possible hallucination was the only one that was 100% factual.

Whoops we had some scrolling bugs with sound, hopefully fixed

And here I thought my being a murder victim was bad.

I looked up the city and year cited by the model for my untimely demise, and it turns out the crime is real, but the real victim was a female sharing my last name, with a middle name loosely resembling my first.

Well, at least he wasn't in the Ep*tein files!

There seems to be some top twenty that rank highly, probably in part due to them being in the files that can't be named!

Straight to jail bud, the AI says you're guilty so it must be true