alt Hacker NewsThe same recovery paths and risk boundaries apply as always.
If anything, these "random number generators" (which they aren't, do not ever use them for that, obviously) are:
- even more cautious,
- more readily take, stash, test and use backups,
- check for and take recovery paths,
- do dry runs,
- consider interruptions,
- mine for more information,
- smoke-test more,
- and are better equipped to do both deeper and wider analyses when encountering problems,
than the typical person.
I'm being very literal about the randomness. The language model outputs a distribution over all possible output tokens, and the chosen output is sampled from that distribution. The process intentionally includes output that is not the top-ranked token according to the model.