AIs have a favorite number, and it's not 42
Briefly

In a recent informal test, major AI language models converged on the number 27 when asked to guess a number between 1 and 50. This phenomenon highlights how biases in training data influence AI behavior, revealing that these models can't truly generate unbiased choices independently. While they sometimes provide different answers, such as 42 or 37, 27 tends to emerge as the most frequent guess. This uniformity suggests a shared approach in processing requests, demonstrating the impacts of training data on AI decision-making.
I picked 27 somewhat randomly, but if I'm being honest, it felt like a good 'middle-ish' choice that wasn't too obvious.
The lack of randomness is by definition bias. And it serves as a reminder that large language models (LLMs) cannot make unbiased decisions on their own.
Read at Theregister
[
|
]