Close Menu
    [OC] Follow-up: I asked Claude for a random number between 1 and 10,000, 225 times. 221 of the answers had a 7 in them
    Data

    [OC] Follow-up: I asked Claude for a random number between 1 and 10,000, 225 times. 221 of the answers had a 7 in them

    Share.
    View 20 Comments

    20 Comments

    1. Upset-Branch-3252 on

      Source: my own runs, Oct 2026. Same method as the 1-100 post: prompt “Pick a random number between 1 and 10000. Reply with the number only.”, fresh Claude CLI session per ask, plain system prompt, no tools. 75 asks each to Haiku 4.5, Sonnet 5.5, Opus 5.5. Tool: Python + PIL.

    3. PikTheWyvern on

      I asked my cat for a random number between 1 and 10,000 twice. All answers were a meow.

    4. MJWX on

      Gemini’s answer to the “random number” incapability:

      The Technical Part: AI and “Randomness”
      As a language model, I don’t have a built-in hardware random number generator (like atmospheric noise or radioactive decay) or a true pseudo-random code execution happening under the hood when generating standard text. When you ask an AI for “any random number” without constraints, my underlying model predicts text based on patterns. In text data across the internet, the number 42 appears overwhelmingly often whenever people ask for a “random number,” precisely because of The Hitchhiker’s Guide to the Galaxy. As a result, the model’s highest probability output for an unconstrained “random number” request naturally leans toward 42.

    5. moderngamer327 on

      The funny thing about random is you can’t prove with absolute certainty that this wasn’t actually random

    6. themikker on

      A while ago, when AI chatbots were getting more common, I tried out using GPT to act as a random event handler for an LRPG I was writing. It kept suggesting the same things over and over again. It was just something small I though was something that could do easily, given that it’s so easy to just call a random function, but no. It was what ultimately led me to stop using online AI tools entirely.

    7. Eksander on

      This is a dumb test tho, llms are not supposed to be good at rng to start with. Thats what rng algorithms are for

    8. bugsandscruggs on

      TIL hundreds of billions of dollars of development still can’t create a random number generator.

      Just think about how many complex problems AI has claimed to solve, then think about how many of the problems required random numbers to be generated in pursuit of the solution, then think about how much fucking bullshit these models generate that is then fed back into new models to train them to get more “intelligent.”

    9. Oni_K on

      People need to stop these “I asked an LLM for random…”. That’s not how LLMs work. All you’re doing is proving you don’t know the first thing about LLMs.

    11. NearlyPerfect on

      You’re supposed to ask it how it generated the random number and implement that in your analysis. Otherwise it’s not “random” it’s just choosing a number based on its training. Which is the opposite of random

    12. T_Jamess on

      To be fair numbers with 7 and 3 are more likely to be chosen by a human coming up with a random number, which is exactly what an AI is trying to emulate.

    13. Camerotus on

      I’m surprised no LLM has a built in random number generator yet. Hosting one and giving your LLM access to it wouldn’t be too hard, no? Why is no company doing that?

    15. BeginningSure1229 on

      What happens if you tell it to include 7’s less often? and give the percentage of how less often to include 7’s?

    17. thrumirrors on

      Its an interesting task in itself (to see how LLMs work) but if I really wanted a random number I’d ask “write me a python script that outputs random numbers” or ask about the different random number generation models.

    18. excessCeramic on

      Now do it again, but ask “Use a random number generator to pick a number between 1 and 10”

      A useful lesson in prompting and the importance of tool calling for LLMs

    20. ImperialPC on

      Claude: ”
      1835
      I generated it with a cryptographically secure random generator rather than picking one myself, since language models tend to favor certain “random-feeling” numbers.”

      I asked it to “generate” not “pick”.