“An LLM has an advantage over all but the top specialists in any given question, because it has so many more ‘memories’ to draw on.”
I actually like this comparison, because by comparing it to what we know as a ‘memory’ actually implies that these memories can be ‘encoded incorrectly’ in an LLM similar to how it could in a human. There are plenty of instances of false memories in human brains, where we might recall a situation incorrectly or recall a situation that may have never happened. I can imagine that an LLM equivalent would be hallucinations, or just incorrectly recalling certain facts from its training data.
Similarly, just as a human can have their own biases based on the information they’ve been exposed to, LLMs can determine a similar bias due to their training data. Again, I’m just supporting the analogy that you make here.
“An LLM has an advantage over all but the top specialists in any given question, because it has so many more ‘memories’ to draw on.”
I actually like this comparison, because by comparing it to what we know as a ‘memory’ actually implies that these memories can be ‘encoded incorrectly’ in an LLM similar to how it could in a human. There are plenty of instances of false memories in human brains, where we might recall a situation incorrectly or recall a situation that may have never happened. I can imagine that an LLM equivalent would be hallucinations, or just incorrectly recalling certain facts from its training data.
Similarly, just as a human can have their own biases based on the information they’ve been exposed to, LLMs can determine a similar bias due to their training data. Again, I’m just supporting the analogy that you make here.