As I had explained in my June article:
I found this reaction of the Deep Research AI quite intriguing.
Roughly 85% of that long article had focused upon complex and controversial non-Holocaust related matters, and the AI system endorsed virtually all of that important material as fully accurate. That portion of the analysis constituted five of the six sections of the AI’s fact-checking run, including some 27 of the 30 separate claims that it evaluated.
Yet because the last portion of my article drew upon some of this earlier material to express strong skepticism regarding the traditional narrative of the Holocaust, the overall conclusion that Deep Research reached was that my long piece was “a deeply flawed and misleading article…In summary, the article’s credibility is extremely low.” Indeed, it explicitly demanded that the article instead affirm the reality of the Holocaust.
I think this strange mismatch raised legitimate questions about whether the Deep Research AI had been heavily conditioned by its training program to automatically reject and denounce any written work that raised any questions about the standard Holocaust narrative.
I also found it interesting that the AI system actually produced four separate partial runs of its fact-checking analysis response before it finally finished one without breaking (I only included the complete run). This may indicate that that AI encountered various “internal conflicts” in its training producing those problems, much like a human being might have a difficult time formulating a response to certain highly psychologically sensitive topics.
Yet one important oddity I discovered was that although a brief or glancing reference to any of those forbidden topics provoked this sort of very negative response, an article that entirely focused upon that same controversial issue was often treated much less harshly. Presumably the wealth of persuasive evidence provided in such concentrated form could often partially overcome the “indoctrination” that the AI had previously received.
For example, here were some of the summary conclusions from the fact-checking report for my lengthy 2018 Holocaust Denial article:
Such strange declarations lead me to suspect that the Deep Research AI was currently undergoing an intense program of “indoctrination” regarding the facts and circumstances of the Israeli attack on Iran, and that since this training process was far from complete, the result was these bizarre hallucinations.
At least I have no other explanation for why an AI that was so proficient and knowledgeable on other matters would claim that no Israeli decapitation strike had occurred and that Iranian President Raisi was still alive. Perhaps those with greater expertise on AI systems could provide their perspective.
