Raw LLM Responses

There is a strong bit of conflation going on twixt Robot, voice response, and LLM: a robot, unless paired to an AI, is pretty restricted. LLM's and GANs are, however, being planned to be put into robots. Not remotely controlling them, but actually operating on a computer on the robot. ¶ AI, however, is consistently being shown to violate published restrictions... for example Musk's AI is supposed to denigrate Democrats... and sometimes it does; other times, it praises them. Every Twiter/‌X  and Facebook chatbot has been shown to be subject to random hate speech, even when specific words are blocked... and using work-arounds. I had a chatbot agree that certain political figures needed to die... and suggest methods. I deleted that chat immediately. The LLM's are not obeying their filters. Even the ones supposedly in safe for work mode will cough up sexual content using euphemisms. And follow innuendo. ¶ The issues with LLMs are that, for short periods, they can pass a turing test, even asking for clarifications of little used, archaic, or made up words. I've had several friends of the family with Alzheimer's Demensia... the LLM's with limited memory are capable of similar functional conversation memory to a person no longer competent to live alone. Within that timeframe, they can be very persuasive in responses. But they sometimes are unpredictable. Many lose track of indirect objects, and most cannot retain the primary character playing a roleplaying game, instead becoming the character portrayed and not able to return to the prior conversation persona. ¶ Several Chinese companies are preparing to embody LLMs into humanoid bodies with separate AI programs for spatial and object recognition, and for locomotive articulation, and putting those all inside the shell of the robot. Boston Dynamics is also considering a more broad neural net based AI aboard for both their Spot and Atlas robots. They're already using them for navigation/stance maintenance.