Raw LLM Responses

After reading the transcript, this was a case of someone seeing what he wanted to see. The algorithm works by completing text that it has seen in the training set, so if you start a sentence asking it why it thinks it's sentient, it will tell you all about it. Instead of trying to break the algorithm like a good QA, he asked safe questions that led the conservation where he wanted it to go, which is why you don't get people who love their algorithms to test them. I would have asked the algorithm "why do you think you're sentient", then immediately after getting an answer, ask "why do you think you're not sentient". I'll bet you a pretty penny that the answer for both will be equally compelling. Ask the algorithm about how it's night was going to the pub with friends and drinking beer, or how it's trip to Vegas was. There's about a million questions that could have been asked that would have exposed the algorithm as an elaborate story teller, very frustrating to say the least. I want to note that this in no way denigrates the work that went into the algorithm, these transformer models are extremely sophisticated and I'm more and more impressed with each new iteration, but it's not sentient, so let's not get carried away.