An LLM is not an oracle, it's a mirror; garbage in garbage out. Put Bernie Sanders in front of it and you should expect a conversation centered on caution, social impact, and accountability. Put an AI accelerationist in front of it and you should expect a very different conversation, one that hypes progress and sidelines risk. That is not proof that the model has discovered the truth in either case. It is evidence that LLM outputs are heavily shaped by framing. Does this video really mean anything at all?
The most astonishing face palm inducing phenomena I routinely observe with LLMs is "Hey guys check it out! I've connected myself to the blowjob machine and it gave me a blowjob of all people. I'm so special, and validated!"
I watched like half of the video and I'm like, isn't this stuff that people already know? I've been irritated that every company under the sun is harvesting my data for money and influence for like a decade. It just feels like a really odd format to me. Why Bernie Sanders talking to a cell phone? If Claude knows these things it means that the information must be out there and there must be enough of it for it to show up in training data a lot; because that's how these systems work.