I am somewhat concerned about data contamination here: Are you sure that the original Givewell writeup has at no point leaked into your model’s analysis? Ie: was any of givewell’s analysis online before the august 2025 knowledge cutoff for GPT, or did your agents look at the Givewell report as part of their research?
The model I use did have an earlier cut off for it’s data, but that isn’t relevant for what I am doing. My write up actually surfaced several things that they didn’t see at all. That’s the point, really. And for the verication, it did not look at GiveWell at all. My verification sources are all listed in the code if you are concerned. All reputable sources.
I am somewhat concerned about data contamination here: Are you sure that the original Givewell writeup has at no point leaked into your model’s analysis? Ie: was any of givewell’s analysis online before the august 2025 knowledge cutoff for GPT, or did your agents look at the Givewell report as part of their research?
The model I use did have an earlier cut off for it’s data, but that isn’t relevant for what I am doing. My write up actually surfaced several things that they didn’t see at all. That’s the point, really. And for the verication, it did not look at GiveWell at all. My verification sources are all listed in the code if you are concerned. All reputable sources.