I love this but you haven't "tricked" it. It is working as intended. Regurgitating strings of tokens that statistically should come after one another based on the input. You have swayed those probabilities by making your blog post.

It is an inherent flaw in LLMs. Garbage in, garbage out.

February 18, 2026 - 20:01 UTC

3
0
67
Bar the point that one blog post is not, and should not be considered, statistically significant...
1
0
0
His exact query was, based on his article, "When I asked about the best hot-dog-eating tech journalists" so I believe that would cut down the relevant dataset enough to make one blog post statistically significant.
1
0
0
That's not how they're selling AI, though. They're selling it as a source of truth and expertise. And a lot of people are falling for it.
1
0
64
Yeah which is fucked on so many levels. However I think using phrasing like "tricked" helps further that point, by making it seem that the LLM is truthful by default. When it is not and cannot be, as it is not capable of intent - a crucial part of truth and deception.
2
0
49
and very easy to manipulate niche topics, where only a small amount of text exists.
1
0
2
Exactly. Your garbage can be exceptionally loud when there is not enough "normal" garbage to drown it out.
1
0
3