|
|
|
|
|
|
by Jack5500
1223 days ago
|
|
|
Some of the "observations" he quotes are most likely made up, I think.
Especially the one by Roman Semenov does seem odd to me.
Either way, reproducibility is a problem with these "jail-breaks" and until we have a LLM, where we can pass a token and get these reproducibile results, I would take all these articles about the "dark side of LLMs" with a grain of salt. |
|
|
- let's reflect on what "made up" means in this context
- you're saying we should doubt the critics of an AI system because their results aren't reproducible. But the creators of this AI system have made it closed, subject to constant change without notice, and to my understanding don't support _any_ reproducible use of it. Should making an system which resists reproducible invocations mean that system is invulnerable to critique? Should we not also view all the system's alleged successes with the same doubt?