What LLMs Still Can't Do
Too Long; Didn't ReadLarge Language Models (LLMs) such as GPT-3 are powerful text generators, but they lack true common sense understanding. Common sense encompasses a broad range of knowledge and abilities, including understanding the physical world, human relationships, cause and effect, making inferences, and applying knowledge to new situations. Hubert L. Dreyfus, in his book "What Computers Still Can't Do," critiqued AI's common sense capabilities, highlighting issues such as the lack of embodied experience, difficulty in understanding context, the symbol-grounding problem, and the absence of human intuition. While modern LLMs represent a different paradigm of AI, the challenge of imbuing machines with true common sense remains, and researchers continue to work on enhancing AI's common sense reasoning.