The Pope appears to understand AI better than Geoffrey Hinton does.
We recorded 9 claims from this post. 3 were checked against evidence. 6 can’t be checked. Read the post at its source.
Claims from this post (9)
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
“Claude’s outputs are the product of a form of mimicry, rather than as a report of genuine internal states.”
What we tested: Claude's outputs come from a form of mimicry and are not reports of genuine internal states.
Assessment: Can’t be checkedSaid
“Consciousness is about internal states; the mimicry, no matter how rich, proves very little.”
What we tested: However rich it is, LLM mimicry of human output proves very little about consciousness, because consciousness is about internal states.
Assessment: SupportedSaid
“Dawkins seems to imagine that since LLMs say things people do, they must be like people, and that simply does not follow.”
What we tested: It does not follow that LLMs are like people just because they say things people say.
What happened: Anthropic reports approximate and exact arithmetic pathways, while Claude's verbal explanation describes the familiar carrying algorithm. This supports the inference that human-like explanatory language alone cannot identify the process producing it. (Anthropic, )
Assessment: MixedSaid
“And the differences are immense; one (the LLM) effectively memorizes the entire internet; the other (the human) builds a mental model through experience with world.”
What we tested: LLMs differ immensely from humans: LLMs effectively memorize the internet, while humans build mental models through experience of the world.
What held: Participants acquired structural knowledge and used it to infer previously unobserved paths and suitable behavior. This supports the human experiential model-building clause, within the study's controlled tasks. (Nature Communications, )
What did not: The study finds structured spatial and temporal representations in Llama-2 across datasets and entity types. This challenges the contrast between LLM memorization and human model-building, without showing that LLM representations equal human mental models. (arXiv, )
Assessment: Can’t be checkedSaid
“You can’t just look at the outputs, without investigating the underlying mechanisms, and conclude that two entities with similar outputs reach those similar outputs by similar means.”
What we tested: One must not look only at the outputs of two entities with similar outputs, without investigating the underlying mechanisms, and conclude that they reach those outputs by similar means.
Assessment: Can’t be checkedSaid
“LLM researchers are NOT creating beings.”
What we tested: LLM researchers are not creating beings.
Assessment: UnresolvedSaid
“They are creating interactive fiction that is trained to predict the language of actual beings.”
What we tested: LLMs are interactive fiction trained to predict the language of actual beings.
Assessment: Can’t be checkedSaid
“And Hinton should know better.”
What we tested: Geoffrey Hinton should know better than to treat what LLM researchers create as beings.
Assessment: Can’t be checkedSaid
“a. That’s the very Eliza illusion I am pointing out. It is possible to fool people into seeing things that aren’t real (e.g., pareidolia, and”
What we tested: Grimes's subjective impression that there is something there in AI is the Eliza illusion Marcus is pointing out, and it is possible to fool people into seeing things that aren't real.