HackerTrans
トップ新着トレンドコメント過去質問紹介求人

cmrx64

2,291 カルマ登録 14 年前
https://github.com/emberian

[email protected]

投稿

Show HN: CakeML-based self-verifying, self-improving system

emberian.github.io
4 ポイント·投稿者 cmrx64·先月·0 コメント

コメント

cmrx64
·昨日·議論
i consider it the root menu of the phone
cmrx64
·9 日前·議論
i got so inspired by reading diaspora this year that i instantly started working on some polisware. cipherclerk operational: https://github.com/emberian/dregg

topical to the conversation, it is fully formally verified in lean (with some UC security reductions done in isabelle). also did this in HOL4 inspired by some work i did with ramana kumar in 2016, on reflective self-verifying self-modifying systems: https://github.com/emberian/svenvs
cmrx64
·先月·議論
it is unlikely that other mosquitos (that don’t carry disease) will fail to fill the niche
cmrx64
·2 か月前·議論
thinking about how to communicate it in a clear way. “the control knobs of what actually makes us different from one another” — you don’t expect one of your kids to be quadripedal. otoh this doesn’t really capture the precise notion either.
cmrx64
·2 か月前·議論
The Electronic Harpsichord, same year. must have been an interesting time.
cmrx64
·2 か月前·議論
LLM API endpoint does do compaction. OpenAI definitely does support serverside compaction, both explicit and automatic, and this is different than what could be implemented purely clientside: https://developers.openai.com/api/docs/guides/compaction (and there was rumors a few months ago on HN about how activation-preserving/latent it is, vs just summarization). Anthropic as well, in beta (new to me): https://platform.claude.com/docs/en/build-with-claude/compac...
cmrx64
·2 か月前·議論
and soon from space? radio engineering breakdown of starlink radar capabilities, it’s a pretty impressive bird if you were designing it only for that: https://youtu.be/jbp3kdJZ1_A
cmrx64
·3 か月前·議論
four figures these days. fits in your hand. nanopore is a revolution. https://nanoporetech.com/products/sequence/minion

there’s youtubers that have videos about doing this in a home wetlab. very achievable. some amateur soil biologists using this to try and sample microdiversity as the planet… humanifies.
cmrx64
·3 か月前·議論
i think of them as tiers of expertise— need to master the basics of structure and form before the robot has the learned representations to competently model user interactions with more fluid instantiations (by downprojecting into the overlearned fixed-semantics)
cmrx64
·3 か月前·議論
claude-opus-4-7
cmrx64
·3 か月前·議論
it used to do this naturally sometimes, quite often in my runtime debugging.
cmrx64
·3 か月前·議論
Dijkstra would shake his head at our folly.
cmrx64
·3 か月前·議論
quite interested; email in bio.
cmrx64
·3 か月前·議論
Those offices themselves were liminal (in sense of ritual) in a way, providing the threshold for the information age to take root.
cmrx64
·4 か月前·議論
this is a pretty bad vfs. there are pure “cap manifest” approaches that don’t pull in decades of cruft semantics. don’t build systems that aren’t objectstore native in 2025 (since this work was initiated in december).
cmrx64
·4 か月前·議論
meh. i’ll keep using scrapinghub.
cmrx64
·5 か月前·議論
https://hellas.ai is building out their category theoretic compiler and protocol for solving this issue
cmrx64
·5 か月前·議論
RIP
cmrx64
·5 か月前·議論
aodv works great and is “actual mesh.” the ‘overlays’ scale in practice and are topology agnostic, the hierarchical bgp mesh underneath isn’t altering the message or memory complexity, we can talk about them as algorithms. there are 10ks node meshes in the real world that use batman, geography-contrained hub and spoke, etc. guifi has 37k nodes in its heterogenous mesh with a batman fork, freifunk (originator of batman) around 40k.

edit to add: what is observed in practice is that gossip protocols can’t coordinate peers without centralizing. this is natural, and an artifact of the logarithmics in the routing protocols. the appropriate thing to do is model routing as a revocable proof system, and information theory explains the centralizing dynamics (problem I worked on 2018-2019) https://eprint.iacr.org/2022/1478 proves the global lower bound is linear (in route updates, when applied to routing, naively quadratic if distributed), and the trick routing protocols add to the game is locality, which yields the logarithmic advantage that when multiplied across the entire network is substantially subquadratic.
cmrx64
·5 か月前·議論
this isn’t a theorem of network science, and is an easily avoided failure mode. plumtree? kad? aodv? you’re wrong :(