← All IntelClip / OtherSuspected reactive patching: companies blacklisting viral failure modes
From Why can't ChatGPT Voice set a timer? | Voice AI expert explains · ≈31:09
“I really thought, okay, they kind of blacklisted beatboxing because of him.”
“I can imagine that they would just like basically have someone who's there watching his videos and then go like, "Well, that's a problem."”
“Like they've probably got someone who's basically generating like 50,000 like training examples, like synthetic examples of what he's been doing.”
What’s in it
- Speculates OpenAI quietly patches ChatGPT after viral quirks go public
- Explains how a beatboxing jailbreak got silently blacklisted post-virality
- Floats theory that OpenAI generates synthetic training data to kill exploits
Clip transcript
what >> Yeah, this one as much as I tried, I just couldn't reproduce it. And um that also made me wonder, is Open AI watching his videos and fixing these things? I'm sure that there are some if and else's somewhere in their code base. For instance, after his beatboxing video, for a long time, I couldn't get chat GPT to beatbox at all. It was just saying, "Sorry, that's not something I want to do." So, I really thought, okay, they kind of blacklisted beatboxing because of him. And maybe this is another thing that they fixed immediately. I don't know. >> Yeah, [snorts] I can imagine that they would just like basically have someone who's there watching his videos and then go like, "Well, that's a problem." Like they've probably got someone who's basically generating like 50,000 like training examples, like synthetic examples of what he's been doing. And then basically go like, "No, don't do that thing." Like and then do all right, all right, all right, shut up on that. >> If you were to build a new company today
Comments
Sign in to comment.
Loading comments…