Criticisms

LLMs:

The claim that LLMs, large language models, have human-like general intelligence is undermined by edge cases, "hallucinations" and other fundamental issues.

But the punditry often claim that any seeming difficulties will be solved by scale - increases in computer speed and storage. As latter-day P.T. Barnum, Sam Altman, says: scale will solve everything. This claim has zero factual basis. Fundamental problems that bedeviled Symbolic AI still haven't been solved for LLMs, including severe semantic difficulties.

The huge statistical databases of LLMs are remarkable, and statistical relationships in them produce responses to incoming streams of electronic binary difference, which, when converted by users' devices into text characters on the users' display screens, seem like responses of a thinking human.

But as regards human-like intelligence, this is smoke and mirrors. The binary differences stored as LLM are created by humans pressing keys inscribed with certain text shapes. What's in the LLM was created by human minds that already knew things, already understood.

Everything in the LLM is derivative of pre-existing human knowledge. LLMs themselves don't understand. The humans who pressed the keys inscribed with text shapes understood. 

Human intelligence evolved to survive in a complex, novel, hostile and changing world. No LLM has done this. For a start they need human-like sensory apparatus plus the algorithms and structures for learning in the human sense of learning from nature. 

The fundamental failures of LLMs seem hard to explain. Some needed concepts probably don't exist. Turing with his neuron-like B-Types started the quest for human-like intelligence in a machine. But he failed to address the most important topic - how the machine could have semantic content in the same sense human minds do.