There Must be Some Misunderstanding
We recorded 8 claims from this post. 5 were checked against evidence. 3 can’t be checked. Read the post at its source.
Claims from this post (8)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: Can’t be checkedSaid
“These systems don’t deeply understand their inputs, contra Dr. Hinton, and as pointed out here regularly.”
What we tested: Large language models do not deeply understand their inputs.
Assessment: SupportedSaid
“Compositionality remains a problem. So does negation.”
What we tested: Compositionality and negation remain problems for large language models.
What happened: The researchers tested multiplication, logic puzzles and dynamic programming. GPT-4's multiplication accuracy declined sharply with problem complexity; step-by-step prompting improved simpler cases without eliminating the harder failures. (Allen Institute for Artificial Intelligence, )
Assessment: SupportedSaid
“Reliability remain a problem.”
What we tested: Reliability remains a problem for large language models.
What happened: Table 3 reports GPT-4 prompt-level accuracy of 76.89% under strict checking and 79.30% under loose checking. The November tests therefore show frequent instruction failures, while also showing that most tested prompts succeeded. (arXiv; Jeffrey Zhou and colleagues, )
Assessment: Can’t be checkedSaid
“You shouldn’t count on these systems.”
What we tested: People should not rely on large language model systems.
Assessment: SupportedSaid
“Scale isn’t everything, and hasn’t and won’t be enough to solve any of the problems I just listed above.”
What we tested: Scaling has not been enough to solve the problems of understanding, compositionality, negation, and reliability in LLMs.
What happened: OpenAI describes GPT-4 as a milestone in scaling deep learning while acknowledging factual and reasoning errors. The same disclosure reports reduced hallucinations and improved factuality, showing progress without complete resolution. (OpenAI, )
Assessment: UnresolvedSaid
“Scale isn’t everything, and hasn’t and won’t be enough to solve any of the problems I just listed above.”
What we tested: Scaling will not be enough to solve the problems of understanding, compositionality, negation, and reliability in LLMs.
Assessment: Can’t be checkedSaid
“We really should consider trying some other approaches to AI.”
What we tested: The AI field should consider approaches to AI other than scaling large language models.
Assessment: UnresolvedSaid
“The question is not whether a clever user with a lot of experience and time to kill can induce a mission to do something correctly, but whether the systems reliably deliver what they are asked for. They don’t.”
What we tested: The large language model systems discussed in the post do not reliably deliver what they are asked for, even though clever, experienced users with time to spare can sometimes induce them to do something correctly.