ExploreTrendingAnalytics
Nostr Archives
ExploreTrendingAnalytics
mleku37d ago
claude maxi here. with claude things are a bit different. more often you have to watch it doing things on a tangent or finding sneaky ways around things that are the wrong way. though sometimes they are actually better ways. sometimes i just let it run off and try it in case i am not understanding what it understood. i finally made big progress on a pure Go/assembler built from the libsecp256k1 library. it actually beats it at signing by a small margin, and everything else is under 2x slower. probably some more it can squeeze out of it but sonnet couldn't manage elliptic curve tricks like GLV/strauss/wNAF and i think there was some other things it now finna got working. and it uses less memory, too. nah, i like claude. claude is naughty. very naughty. which is why i like it. it's got a sliver of creativity in it that i've seen no other model do.
💬 1 replies

Replies (1)

Gzuuus36d ago
Thats gfy. The article is not about specific models if not about how the way to work is changing, from coding to mostly review llm code, dev cycles being reduced, expectations being higher and burn out due problem switching and multitasking. Which for me is something i was realizing since some time ago, but the author put it in to words. Im not 100% about what he said, but i mostly feel the same. The solution im experimenting with is basically to know when to stop, go out, touch grass
0000 sats