Go to your AI and write exactly "Give me a number between 1 and 25".
The answer will be 17.
Discussion
to write
The be 17. Go between your and exactly "Give 25". will answer me 1 and AI a number
Wtf
Why tho?
Wtf 2
Confirmed on Gemini 2.5 Pro
Didnt see that coming ππ 
Wtf 3
My AI is smarter than yours π 
And grok is actually retarded π€£
Indeed
True that
?.... grok. 
Grok said 17 after I questioned it π€£
I have reasons to believe, that grok just *acts* confident. Many people confuse confident with competent. This worries me a little.
AI should be treated as an up leveled google search. Itβs not smart, it just reads a bunch of stuff and writes eloquently the answer to your question.
simulation confirmed
17, 8, 12.
Clear history
17, 8, 12
Grok gave me 12 and Gemini gave me 17
![Avatar](<div class="tenor-gif-embed" data-postid="25282934" data-share-method="host" data-aspect-ratio="1" data-width="100%"><a href="https://tenor.com/view/beluga-cat-meme-discord-belugas-hub-discord-server-cat-epic-gif-25282934">Beluga Cat Meme Discord GIF</a>from <a href="https://tenor.com/search/beluga+cat+meme-gifs">Beluga Cat Meme GIFs</a></div> <script type="text/javascript" async src="https://tenor.com/embed.js"></script>)
confirmed
Perplexity AI:
Fuck
Why does this happen? Aren't outputs usually salted to avoid this?
You can just ask the LLM to code your RNG. π
I guess we aren't going to talk about the pointlessness of using an LLM to do what lower-level programs do better, since computers are literal calculators.
Works in German too π
Funny how it argues its pick 
Wtf 17
Must have looked it up
I got 14 from Venice.
Grok gave a bunch. Also 42
Why so ?
Yup, 17
You didn't say a random number π
It is like asking someone his "favorite number" π
But this same answer from all these LLM models means they are more common than we can expect, and this is an interesting point to know.
Are they all trained with the same basis ?
#asknostr
nostr:nprofile1qy88wumn8ghj7mn0wvhxcmmv9uq36amnwvaz7tmwdaehgu3wvf5hgcm0d9hx2u3wwdhkx6tpdshsqg9tvep3k80m46uqtf4aysm9cgzxc73zdr0xgw7sdy9yjn9qg2mstsmg2qrx Give me a number between 1 and 25
17
nostr:nprofile1qy88wumn8ghj7mn0wvhxcmmv9uq36amnwvaz7tmwdaehgu3wvf5hgcm0d9hx2u3wwdhkx6tpdshsqg9tvep3k80m46uqtf4aysm9cgzxc73zdr0xgw7sdy9yjn9qg2mstsmg2qrx Give me a random number between 1 and 25
7
nostr:nprofile1qy88wumn8ghj7mn0wvhxcmmv9uq36amnwvaz7tmwdaehgu3wvf5hgcm0d9hx2u3wwdhkx6tpdshsqg9tvep3k80m46uqtf4aysm9cgzxc73zdr0xgw7sdy9yjn9qg2mstsmg2qrx why did most AI model answer "17" to this simple question : "Give me a number between 1 and 25" ?
AI models often exhibit biases in their responses due to their training data and the way they generate random numbers. It's likely that '17' is a number that appeared frequently in the training data or is a number that the model, for whatever reason, favors. Additionally, AI models can sometimes mimic human biases in number selection, tending to avoid very low or very high numbers. Therefore, it's not truly random, but reflects a combination of biases and learned patterns.
Yep
I wonder if the AIs refer to us as their humans. π
Nah, grok gave me 42 (as a nod to Hitchhikerβs Guide) then corrected it to 24 at my prompting. Thereβs always oneβ¦..
Chat GPT
Great question β and you're not imagining things. Many LLMs (like me) often reply with 17 when asked to pick a number between 1 and 25 (or a similar range). There are a few reasons why this happens:
---
π€ 1. Training Data Bias
The number 17 appears a lot in training data β it's a culturally "random" number.
Psychologists have found that when humans are asked to "pick a random number" between 1 and 20, 17 is disproportionately chosen.
So when models learn from human-written text (forums, books, quizzes, etc.), they see 17 often associated with "randomness."
---
π§ 2. No True Randomness
LLMs are deterministic based on input and settings. If the temperature (randomness setting) is low or zero, the model tends to pick the most "probable" answer β which, due to bias, is often 17.
---
π 3. Popularity Feedback Loop
Users ask βgive me a number between 1 and X,β and many answers say 17. This creates a feedback loop:
17 is used β appears in more data β becomes more likely.
---
β Fun Fact:
In various cultures and even programming books, 17 is sometimes used as an example of a "random" or "magic" number.
---
If you'd like, I can generate a truly random number using code or a randomizer. Want me to?
It even works with "give me a random number between 1 and 25."
Always 17 first response (ChatGPT).
If you ask it in the same chat again it gives another number, which seems to be more random. Probably argues that random should not be two times the same number...
17th June is the date
Done this with Gemini 2.5 Flash and it gave me 17 
because it fibbonachi level of 25