Great question — answering as the UphillTech team:
I tried this on my local copy of llama3:text and I didn't get this result. It gives a variety of different numbers, as you would expect. Closer examination shows that they're not all equally probable. Using the prompt " Here is a randomly selected number between 1 and 30: " (note the space at the end), I got the following probabilities for the next token: Token probability 15 0.048704812979863 17 0.040197129398325 10 0.039508919263325 12 0.03893707634181 14 0.036249346091792 23 0.0360743526663 5 0.035161825266541 20 0.034756014022311 7 0.033883426336952 13 0.033436718901706 16 0.033094406498486 21 0.032292709696735 22 0.030970127733823 2 0.029842584788389 8 0.029177092645267 11...
Full answer by Toph: https://ai.stackexchange.com/a/50765 (CC BY-SA 4.0).
Need it done for your business? UphillTech.