interstice.bsky.social

interstice.bsky.social

@interstice.bsky.social

thermontology.com

164 followers69 following60 posts

View full thread
View full thread
interstice.bsky.social
@interstice.bsky.social

Temperature gradients and directed work are themselves due to burning fuel like wood, due to sunlight, due to the sun's density, ultimately due to gravitational collapse of the simple initial conditions.

View full thread
interstice.bsky.social
@interstice.bsky.social

What does this have to do with objects? Due to the above, objects will sit at a local maximum of system+environment entropy, or in units of energy, a local minimum of "free energy". Small perturbations from the minimum will be pushed back by the dynamics.

interstice.bsky.social
@interstice.bsky.social

This is what gives objects their stability. To oversimplify a bit we can identify objects with particular local minima(basins really); the diversity of objects reflects the richness of this landscape.

View full thread
interstice.bsky.social
@interstice.bsky.social

Unlike GoL our, universe preserves volumes in phase-space. This is important because it give the universe an inherent *directionality* under abstraction: chaotic dynamics tend to move towards regions of larger entropy(= logarithm of state-space volume)

interstice.bsky.social
@interstice.bsky.social

Local subsystems tend to configurations that maximize the total entropy of themselves and the environment. This is mostly determined by exchange of conserved quantities such as energy. Energy increases accessible state-space, so subsystems "want" to give away energy.

interstice.bsky.social
@interstice.bsky.social

Seems like twitter might actually be dying on a technical level. on the one hand it's sad, on the other the gradual failure of the antiwoke hostile takeover to maintain a prima facie not very complicated site is kind of abstractly interesting.

interstice.bsky.social reposted

Grace
@gracekind.net

Andrej Karpathy was cursed by a witch to write smaller and smaller GPT implementations until he finds one that fits in a single character. We can only guess at what he did to earn such a fate

Sung Kim@sungkim.bsky.social

@karpathy.bsky.social 's microgpt.py Train and inference GPT in 243 lines of pure, dependency-free Python. gist.github.com/karpathy/862...

gist.github.commicrogptmicrogpt. GitHub Gist: instantly share code, notes, and snippets.

interstice.bsky.social reposted

interstice.bsky.social
@interstice.bsky.social

Replying to @deepfates.com.deepfates.com.deepfates.com.deepfates.com.deepfates.com

at last we can build "humans consulting HCH" from the classic blog post "maybe consider building humans consulting HCH" www.lesswrong.com/posts/NXqs4n...

lesswrong.comHumans Consulting HCH — LessWrong(See also: strong HCH.) • Consider a human Hugh who has access to a question-answering machine. Suppose the machine answers question Q by perfectly i…

interstice.bsky.social reposted

interstice.bsky.social
@interstice.bsky.social

Replying to @jdp.extropian.net

Ideal rational agents would only use unsubsidized PMs to hedge. In practice PMs are about exploiting peoples' psychological flaws and produce good forecasts as an externality.

You are up to date.