Angry AI.

Video Description

Tesla's Optimus robot, Elon Musk and the AI LaMDA. https://brilliant.org/digitalengine - a great place to learn about AI and STEM subjects. You can get started for free and the first 200 people will get 20% off a premium annual subscription. Thanks to Brilliant for sponsoring this video. The AI interviews are with GPT-3 and LaMDA, with Synthesia avatars. We never change the AI's words. I have saved the OpenAI chat session to help them analyse the situation and there's a link to the chat records below. I've noticed some people asking if this is real and I can understand this. You can talk to the AI yourself via OpenAI, or watch similar AI interviews on channels like Dr Alan Thompson (who advises governments), and I've posted the AI chat records below (I never change the AI's words). To avoid any doubt, the link now also includes a video of the chat and a copy of the code. It feels like when Boston Dynamics introduced their robots and people thought they were CGI. AI's moving at an incredible pace and AI safety needs to catch up. Please don't feel anxious about this - the AI in this video obviously isn't dangerous (GPT-3 isn't conscious). Some experts use scary videos like 'slaughterbots' to try and get the message across. Others stick to academic discussion and tend to be ignored. I'm never sure of the right balance. I tried to calm anxiety by using a less threatening avatar, stressing that the AI can't really feel angry, and including some jokes. I'm optimistic that the future of AI will be great (if we're careful). Sources: Here are the records for the GPT-3 chat (screenshots and a video to avoid any doubt). I've marked the words from Elon Musk and Ameca on the first page (which I gave the AI to respond to in the previous video): https://www.dropbox.com/sh/82iwek5rnofmrn2/AADM4WOJkjREhR_yMOPHFIF4a?dl=0 Tesla's AI day 2, introducing the Tesla Optimus robot: https://www.youtube.com/watch?v=ODSJsviD_SU&t=1198s Researchers from Oxford University and DeepMind on AI risks: https://onlinelibrary.wiley.com/doi/10.1002/aaai.12064 Robotic Navigation with Large Pre-Trained Models of Language, Vision, and Action: https://arxiv.org/abs/2207.04429