AIs and LLMs are extremely smart and incredibly stupid

Computer scientist Yejin Choi agrees that AI and LLMs can be very smart, but also incredibly stupid. And that stupidity can be easily identified: AI and LLMs lack common sense.

In fact, instilling/programming/training AI for common sense turns out to be very difficult–some of her colleagues say she’s wasting her time trying. Instead, they believe that LLMs will fill in the gap of missing common sense with more training.

But Choi offers the analogy: You don’t get to the moon by repetitively making your buildings an inch taller; you won’t get common sense in AI by expanding it’s training. Something else is required.