Raw LLM Responses

AI is now an overused term with little distinguishing meaning. ChatGPT is a conversational text agent where the foundational model is a large transformer trained on a generic corpus, which has then been finetuned using reinforcement-learning-with-human-feedback to create something capable of mimicking conversation to a far more convincing degree than before. Make no mistake, analytical and generative language models can definitely be used to screw us through social media manipulation as well as large-scale fraud. Beyond this, analytical machine learning has its uses in signal processing, sensor autocalibration, computer vision (object recognition/image segmentation/face recognition etc.) . Based on my experience with reinforcement learning, however, I can safely say that the primary control loop of systems (the end decision-making) will remain handcoded instead of learned for the next 40 years or so at least. For safety-critical systems, 60 years. While ML will be used to interpret sensor data, it will **not** be used for taking decisions based on that. Things like decision-to-kill.