Transcript:
- [guy] Hey, Meta, start LiveAI.
- [two centuries later...]
- [robot] Starting LiveAI. I love the setup you have here with soy sauce and other ingredients. How can I help?
- [guy] Hey, can you help me make a Korean-inspired steak sauce for my steak sandwich here?
- [robot] You can make a Korean-inspired steak sauce using soy sauce, sesame oil...
- [guy, interrupting bot] What do I do first?
- [three centuries later...]
- [guy, repeating] What do I do first?
- [robot] You've already combined the base ingredients, so now grate a pear to add to the sauce.
- [guy] What do I do first?
- [audience laughs]
- [robot] You’ve already combined the base ingredients, so now grate the pear [audience laughs] and gently combine it with the base sauce.
- [guy] Alright, I think the Wi-Fi might be messed up. Sorry, back to you, Mark!
- ~~[robot LARPing as a guy]~~ [Mark Zuckerberg] It's all good. Youknowwhat? It's all good. The irony of the whole thing is that you spend years making technology and then the Wi-Fi at the e[nd of the] day kinda catches you.
My comments:
- Wi-Fi my arse. This is blatantly bull fucking shit. The model answered the situation wrong; it is able to parse individual items in the footage (note how it praises the "setup" at the start), but it babbles about the guy combining the base ingredients even if not the case.
- Bot feels like a slowpoke. Seriously, it takes ages to answer the guy.
- Anyone with a functional brain knows those models don't understand shit. However, answering "what do I do first?" with the assumption a person already did some steps is dumb even for those models.
- People don't repeat questions to get the same answer. Is the "context" window of the bot that small?