When I asked the generative AI a question about Kakuei Tanaka, it gave me a ridiculous (obsolete) answer.
When I wrote an article about Mr. Kakuei Tanaka's mansion, I searched for when he started living there, but I couldn't find any information. So I asked the generation AI.
The source for this "mid-1950s" is the Asahi Shimbun "Tensei Jingo" dated January 10, 2024. Indeed, there is such a description of Mr. Tanaka's residence.
I thought COPILOT was reading the Tensei Jingo on the Asahi Shimbun website, but that wasn't the case. The site that COPILOT used as a source was not the Asahi Shimbun site, but a personal blog. The entire Tensei Jingo article was copied there.
In other words, COPILOT uses information obtained from sites where third parties have copied other people's articles without permission. This is very bad.
The Asahi Shimbun website requires a fee. Collecting information for free from a site whose content has been copied without permission is like reading manga from a pirated site. Also, even if the blog author intentionally or accidentally rewrites some part of the copy, readers will not notice.
So, I asked COPILOT, "The article site you are referring to copies the Asahi Shimbun's Tenseijingo. Do you also collect information from such sites?"
In the end, as I wrote before, the accuracy of the generated AI's answer depends on the information provided by the operator.
Another question I asked the generated AI was, "Was Kakuei Tanaka subpoenaed as a witness in the Lockheed incident?"
As I wrote earlier, Mr. Tanaka has not been subpoenaed as a witness. The site cited by COPILOT as a source contains such expressions.
COPILOT could not read this sentences correctly. Of course, COPILOT probably hasn't even read the minutes of the National Assembly, which is more reliable.
COPILOT is not the only generative AI that gives incorrect answers. I asked Gemini "Did Mr. Kakuei Tanaka say, "I don't remember"? ".
The generation AI is probably being fed poor training data. However, if more and more people accept the incorrect answers provided by the generative AI, it will remain as an established theory.