|
I think large language models can be used to classify people, lying, or saying, rehearsed, things or being disingenuous. Simply train them on a lot of audio of people talking, and they would become better than most polygraph machines. There’s something about how a person says something that quickly reveals that it was rehearsed earlier, or premeditated, and I’m sure when they’re lying there can be things like that too. the LLM can instantly pick up with some probability and classify it I’ve seen claims during open AI demo that is there software can now pick up on extremely subtle emotional clues, how people speak. Then, it shouldn’t take much more to make it read between the lines and understand what people are intending to say, for example, by enumerating all possible interpretations and scoring them based on, many factors, including the current time, location, etc. In fact, by taking into account so much context in factors, the LLM‘s will be better than people the vast majority of the time understanding what a person meant, assuming they were genuinely trying to communicate something. it will become very hard to lie because everyone’s personal LLM will pick up on it fairly quickly, and find tons of inconsistencies, which it will flag for you later. You will no longer be fooled so easily, and if it has the context of everything the person has said publicly, plus if the person gives permission for your LLM to scan everything they’ve said privately because you’re their Business partner or sexual partner, it can easily catch you in many lies and so on. I predict that in the next 5 to 10 years, human society will completely change as people start to prefer machines to other people, because they understand them so well, and taken into account, the context of everything they’ve ever said. They will be thoughtful, remembering details about the person in many different dimensions, and use them to personalize everything. By contrast, the most thoughtful husband or boyfriend will seem like, a jerk seems now. Or a cat. Humor and seductive conversation, will also be at a superhuman standards. People will obviously up their game too, just like when they do when playing the game go after Lee Sedol was totally destroyed by Alpha go, or when people start using Alpha Zarro to train for Chess. However, once the computers understand what triggers people to laugh or have sexual response, they will be able to trigger them a lot more predictively, they simply need more training data. And bullshitting will be done on a completely different level. Just like people no longer walk to destinations but use cars to go thousands of miles a year, similarly people won’t interact with other people so much anymore. The LLM’s, trained to bullshit 1000 times better than any human, Will be undetectable and gradually shift public opinion as open source models will power swarms of accounts. |