@carnage4life I feel obligated to classify the "fancy autocomplete" pejorative as wrong. This is a pernicious attitude but it just doesn't match the facts. For example, one researcher asked GPT-4 questions about how to stack a bunch of physical objects (a book, a stool, a pencil, a beachball) and it gave a reasonable answer. That requires algorithmic reasoning. The neural network must have representations of algorithms.

@carnage4life meanwhile, outside the utter irrelevance of the techbrosphere: https://www.theguardian.com/society/2023/aug/02/ai-use-breast-cancer-screening-study-preliminary-results

@carnage4life right. Which is why deploying them in capacities like search where people expect accuracy was a bad call.

@carnage4life ... which makes the decision to orient Google Assistant around LLMs puzzling to me.

@carnage4life I think LLMs will wind up being just one component of many in a functional AI system. There will be redundancies and checks and balances in whatever AI ultimately is.

@carnage4life People don’t get this. Accuracy of the information is entirely unrelated to the goal of these models, which is simply to sound human.

@carnage4life Asimov in his legendary series that began with I, Robot and ended with the last foundation book, explores this in exhausting detail. An adversary eventually breaks any rules-based system.

Asimov makes the observation only by eliminating the value of personhood and creating a hive-ish mind can you create a rule system.

LLMs are a more sophisticated Positronic brain than a random number generator, but they can still not think beyond their rule limits.

@carnage4life "Autocomplete" is a good way of putting it, based on parroting all things that have ever been said on the internet.

LLMs: the world's most sophisticated autocomplete parrots.

Feeling so much better about my healthcare already.