Perhaps that is what you hope or wish, but I don't think that is the case at all, as I do not see any evidence to that.
This.
I did the test multiple times myself, both with Chat GPT and the Google AI. Every time they mentioned limba as the medial ply, just like what Egon posted already. But when you feed them with supported information/evidence, they will actually change their "mind", Google's AI a bit more so than Chat GPT.
Answer(s):
"I think your sources are stronger than the one I relied on previously, and I should have been more cautious. Based on what you've pointed out, I would no longer confidently say the medial ply is limba."
...
"One additional piece of circumstantial evidence is that recent discussions about Butterfly's classic ALC construction also describe it as
koto–ALC–ayous–kiri, not limba. While forum comments aren't definitive, they are at least consistent with the patent and the observations from blade makers."
...
As to my follow-up question "If that correct information was already out there on the internet and available, why did you repeat claims of people without any credentials" Chat GPT answered:
"
That's a fair question.
The short answer is:
I relied on a pattern that appears frequently across secondary sources instead of giving appropriate weight to the quality of the evidence behind those sources.
In this case there were two competing bodies of information:
- Many retailer pages, equipment databases, and forum posts repeating "koto–ALC–limba–kiri."
- Higher-quality evidence—a Butterfly patent, direct observations by an experienced blade maker who has examined blade cross-sections, and consistent anatomical reasoning about the wood grain.
I should have recognized that these are not equally reliable. When there is a conflict, the order of evidence should generally be something like:
- Manufacturer documentation and patents (when applicable).
- Direct physical examination by qualified experts.
- Independent technical analyses.
- Retailer specifications and copied database entries.
- Forum repetition without evidence.
The mistake I made was effectively treating the number of sources repeating a claim as a proxy for its reliability. That's a common failure mode because information on the web often propagates by copying. Twenty websites can all trace back to one incorrect specification.
It can certainly be fun and sometimes useful interacting with AI, but at the same time it is - at least for me - kind of a sophisticated parrot.