maybe that's "non-neurally represented information"

yeah! though i'm iffy on their concept of "information". their example of socially transmittable body attributes - for sure part of culture, i dunno information.

sounds like bsky.app/profile/tedu...

i'm used to something like

"Culture is information capable of affecting individuals’ behavior that they acquire from other members of their species by teaching, imitation, and other forms of social transmission"

boyd&richerson 2005 ia801200.us.archive.org/33/items/The...

I thought I was following this stuff but still had Common Pile and Common Corpus confused until now... I guess their arxiv papers were posted within days of each other!

a systems neuroscientist once mused to me, "maybe we'll never understand the brain". in not an entirely bleak way - they were and are devoted to working on it! - but it's worth being honest about the challenges.

the example i always use is, that (or at least, I think..) traffic engineering uses agent-based model simulations to help infer changes to traffic under counterfactual situations. to be useful there, certain aspects of agent cog/reasoning/decision-making are crucial, and many are not.

i feel like more clarity is needed on the intended uses of the social simulations, to know what capabilities are needed, at what accuracies, at the agent level. i felt similarly about older discussions on ABMs, fwiw (maybe i just don't get what's going on)

i've had exactly this problem in the past - thanks!

see you at HAVI!

good book, assigned in my soc 1
en.wikipedia.org/wiki/Street_...

ah i mean all the uses for stuff that's not (academic) research. like Fig 3 here (2024?). there's.. lots, i guess. arxiv.org/pdf/2309.11998

so there is some magic and i think it's understandable we find this interesting!, but (as these papers find, or i guess we all have) it's not robust or consistently generalizable in ways we can predict, so not trustworthy at the level we'd like for, say, high quality social science.

i guess the new thing is there's now a level of casual, truly zero-shot use with no evaluations or test sets, which is genuinely useful in some settings, and they're settings where in the past annotation was a prerequisite

similar?- we observed tesseract working worse than gemini ocr for book scans, evaluating a downstream relation extraction task. arxiv.org/pdf/2505.10798
we didn't put the tesseract-based numbers in the paper; would have been new rows in Table 2 with a diff values under the "OCR" column