AnthonyA potentially interesting question: how much would the appearance of sentience or intelligence that LLMs can generate for some users explode if they were forced to have deterministic output?
In principle you could add a single "freeze the random seed" toggle to any of the major chatbots, and with that setting toggled on they would always return precisely the same output for a given input. Organisms and by extension humans cannot behave like this---no matter how stereotyped an organism's response may seem, it always differs, in however small a way, from a previous response---and the LLM's illusion should immediately be obvious by contrast. But, perhaps more interestingly for the folks who do think LLMs exhibit some form of sentience or intelligence: are we really meant to believe that a random number generator is the source of sentience or intelligence? You could hook up a random number generator to a machine that is otherwise deterministic and clearly not sentient or intelligent, and it suddenly becomes so? How do you explain that?
#AI #GenAI #GenerativeAI #LLMs
In principle you could add a single "freeze the random seed" toggle to any of the major chatbots, and with that setting toggled on they would always return precisely the same output for a given input. Organisms and by extension humans cannot behave like this---no matter how stereotyped an organism's response may seem, it always differs, in however small a way, from a previous response---and the LLM's illusion should immediately be obvious by contrast. But, perhaps more interestingly for the folks who do think LLMs exhibit some form of sentience or intelligence: are we really meant to believe that a random number generator is the source of sentience or intelligence? You could hook up a random number generator to a machine that is otherwise deterministic and clearly not sentient or intelligent, and it suddenly becomes so? How do you explain that?
#AI #GenAI #GenerativeAI #LLMs
pbog