Interesting how much a child’s growth slows before puberty. 90% of brain volume is present at age 5. So there's a long period in which the brain is mostly developed, where the child doesn’t physically grow that much.
Is there a known explanation of this?
My first thought is that most learning benefits from a big brain but not from a bigger body. E.g. A bigger body is irrelevant for learning language, socialization, and general knowledge. Moreover, it’s probably easier to learn motor skills with a smaller body because you can fall over more times (mass scales cubically) and it’s easier for parents to control kids with smaller bodies while they are still learning impulse control. Brains and bodies both have substantial metabolic costs but the costs of the post-puberty body are not worth it during this period.
Some recent papers claim that adolescent growth spurts are also found in bonobos and other primates. So this growth pattern is not unique to humans — though I assume humans have the longest period in the slow-growth mode.
Conversation
actually this is completely wrong, just because the size of the brain doesn't change much doesn't mean there isn't a lot of neuronal pruning that happens in development(the sparse MOE/lottery ticket hypothesis analogy here is a funny one)
Peptide Sciences™ brings you “Ageless” , a detailed exploration of aging's biological underpinnings. Understand the 14 hallmarks of aging and the research aimed at slowing its pace. This eBook is essential for those interested in the science of aging. Download today.
Discover more
Sourced from across X
I've always believed I have better-than-average intuition. But this year+ of shilling DeepSeek while most people thought of it as, at best, a cute gimmick, "oh, The Whale lol, anon has a thing", has nontrivially bolstered my self-confidence
Quote
Teortaxes
@teortaxesTex
Replying to @VictorTaelin
I first thought "wait a minute, these guys are going to be big" when their first model, DeepSeek-Coder, a basic 33B Llama-like, came out. They've been consistently prioritizing coding since the start.
Lotta people thought this in the early days of ARC, so lotta people tried using vision models.
They all failed badly.
Quote
roon
@tszzl
arc is hard due to perception rather than reasoning -> seems clear and shut x.com/mikb0b/status/…
gifted my high school sister chatgpt pro for christmas
in the last couple hours she went from 0 to legit results on an interp project exploring emergence of position dependent neurons in rnns trained for path integration (she’s interested in entorhinal grid cells)