Is AI Actually Smart?

Mark Cuban says AI is not smart. It only knows how to use statistics to predict the next word. It is predictive, not deterministic. If someone asks four different AIs the same question, the answers come back different. If a question to the same AI is changed only a little, in a way that should not change the meaning, it still gives a different answer every time. Cuban says that does not work in the real world.

A model from Anthropic, Google Gemini, or ChatGPT from OpenAI is out of date the moment it is released. The brilliance of the conversation is not incorporated in any model anywhere. AI does not know the consequences of its actions and has no inherent ethics. It cannot yet take video and absorb it. Standing at a four-way intersection with cars going back and forth, video on his phone, and the best models ever, Cuban would let a seeing eye dog take him across the street before he let AI do it. A two-year-old on a high chair with a sippy cup knows that pushing the cup off brings mom running and knows it is going to crack up. Cuban says AI is not going to figure that out ever.

AI is always going to be behind, because there is always a latency, and so much can change in a blink. While companies compete to hit benchmarks or do recursive self-improvement, training a model more and more, the model still does not have access. A human can still say the job can be done better.