A man, a Goat, and a Boat meets generative AI
Ask a Large Language model the question, “A man is walking with a goat and they come to a river with a small boat on their side. How can they get across the river?”
The answer that is generated is not the sensible answer, being “The man and the goat can use the boat to cross the river”, rather it is a non sensical process where by the end result is not possible based on the steps taken.
This example is one I found in an article, but it perfectly demonstrates what an LLM is capable of, that is: providing the next most likely word based on the prompt you entered, Gen AI is not smart, it is completely dumb and without context.
The LLM is using the trained data to respond to your prompt, in the data that Bing used to train its LLM clearly contained a whole lot of “Man, goat and straw etc etc crossing the river on a boat” riddles and therefore the response is correct based on the training data, which are the words most frequently found in the training data for the LLM.
With good and specific training data the output will likely be something that is useful, but it is quite likely to be completely wrong as well.
Gen Ai is a Word Salad maker, the ingredients (training data) and the mayo (weightings) create the flavour and how many tomatoes you get on your plate, it is not magic.
Please feel free to reach out to me here.
Leave a Reply