That is the point. Training an LLM on the entire internet never will be reliable, apart of the huge energy waste, not the same as training an LLM on specific tasks in science, medicine, biology, etc., with this they can turn in very usefull tools, as shown, presenting results in hours or minutes in investigations which in traditional way would have least years. AI algorrithm are very efficient in specific tasks, since the first chess computers which roasted even world champions.
It was inevitable, when you need to train GPT on the entirety of the internet and the internet is becoming more and more AI hallucinations.
That is the point. Training an LLM on the entire internet never will be reliable, apart of the huge energy waste, not the same as training an LLM on specific tasks in science, medicine, biology, etc., with this they can turn in very usefull tools, as shown, presenting results in hours or minutes in investigations which in traditional way would have least years. AI algorrithm are very efficient in specific tasks, since the first chess computers which roasted even world champions.
Those MLs don’t automate anything though, they increase output but also increase cost. The AI bubble is about reducing costs by reducing head count.