- Nov 26, 2019
- 17,858
- 9,638
- 51
- Country
- United States
- Gender
- Male
- Faith
- Generic Orthodox Christian
- Marital Status
- Celibate
I fed your argument to chatGPT 5.6 Sol, the latest frontier model, and it replied as follows (and to be clear all the following words are from it, not from me):
I understand that ChatGPT agreed that your theory was plausible. The problem is that an LLM agreeing with a proposed explanation is not independent evidence that the explanation is correct.
ChatGPT is very good at evaluating whether a causal story is coherent. It is also capable of confidently endorsing incorrect descriptions of its own operation, especially when the user supplies a plausible theory and asks whether it fits. I have personally seen ChatGPT invent nonexistent features of its Python environment, so I would not treat its self-description as authoritative merely because it sounds technical.
To establish your theory, one would need evidence such as:
- the earliest articles or social-media posts making the Achilles claim;
- evidence that those claims were based on inference from the Hades scene rather than a false production report;
- examples of particular LLMs producing the claim before it appeared in searchable sources;
- and some basis for distinguishing web retrieval from hallucination or pretrained association.
There is also an important technical distinction here. A model using web search may repeat a false claim because current sources report it. A model without web access may generate Achilles because that character is strongly associated with Odysseus in Hades. Those are different failure modes, and simply saying “ChatGPT agreed” does not tell us which one occurred.
So my objection is not that your theory is impossible. It is that ChatGPT’s agreement cannot serve as proof of the theory, because the reliability of ChatGPT is the very question under discussion.
Upvote
0
