~6:00 He talks about how the exchange of free information will tend toward a monopoly because of mathematics. 6:57 "Instead of decentralization, you end up with hypercentralization."

~8:30 He talks about how the limits of language, itself, and how those limits make self-regulating LLM AI challenging. (8:58) "Words are actually more ambiguous than we think they are. We go through life thinking words are precise, but they're not. Which is actually the reason large models work, because they're a statistical thing in the first place."

9:15 "But, there's another way that does work. Anytime there's an output from AI, of course in some small way it relies on millions or billions of examples from humanity at large...but in any PARTICULAR example of an output, it won't. In a particular example, there'll be a few, a small number, maybe a dozen." He adds that if you can determine WHO contributed to an AI result, you can incentivize folks to put in new data that makes the results better and compensate folks for their contributions. (This introduces his notion of Data Dignity.)

11:16 Data Dignity

https://tinyurl.com/ChatGPTresources

https://www.tiktok.com/@checkers80/video/7419146844518796590 ("Questions for the 'sentient' AI PART 1")

• A study in Consciousness and Cognition finds participants remember words better when they draw them instead of writing them. Across repeated tasks, sketchers consistently beat writers on recall tests.

• Researchers used multidimensional experience sampling to examine thought quality and focus. They observed drawing spark richer visual, motor, and spatial processing, while writing remained linear and verbal.

• The article argues that large language models bypass the messy, generative phase of creation, leaving users to merely tweak finished text. When AI handles the making, the author warns the deeper cognitive work that cements knowledge stays with the machine, not the mind.