Just started, and two made it to be block list. This is going well! :-D
Any conversation
there's no fundamental physical limit that prevents silicon from developing taste and judgement, but language is a lossy compression of reality.
taste comes from experience, from the shadowy mass of episodic memory, which swirls and convects at the boundary of consciousness. the ideas recombine into new ones in the nuclear reactor of the hippocampus.
but no matter how imbued with meaning and expertise is a handmade wardrobe we still get our shit from ikea
ai coding feels like it has reached diminishing returns for me. luna is my default orchestrator, routing work by complexity. in practice, usage lands around 16% luna, 48% sol, 30% astra, and 6% terra. at this point, i care less about each new frontier release. the system already finds a capable model for the job. the frontier is moving beyond my day-to-day work, so iโm happy to settle into predictable models for a while.
setting up some of my workflows based on this writeup - ludwigabap.com/โฆ/on-my-personal-โฆof-the-tools-i-use/
Particularly a big fan of zellij amongst others
Also venturing into having agents enable more of my work - let us see how that goes.
Ikea has rules for naming products (so in theory, you could figure out what it is from the name) -> www.ikea.com/โฆ/6f564c4d-2ccc-46de-โฆ545a3948dc79.html
A thing I've been doing lately is trying to come to the same conclusions others have without looking their the direct techniques and code. Last night I tried to figure out why people do vector based retrieval for RAG and ended up running these experiments: n0tls.com/โฆ/2026-09-03-i-think-โฆrag-with-grep.html
I'm still going to do an explore agent.
ไธญๅฝ็ๅผๆบๆจกๅๅจๆ้ซๆๅฟ็ไธปๅฏผไธ๏ผๅจ้ข่ฎญ็ปๆฐๆฎ้ถๆฎต๏ผๅฐฑ็ญ้คๅคง้ๅฏนๅ
ถไธๅฉ็ๅ
ๅฎน๏ผsft้ถๆฎตๆๅปบ็ๆฐๆฎๆบๅ
ฅๅคง้ๆฟๅ
ถ่พฉๆค็ๅ
ๅฎนใๆฅ้ไธไธช็บ ๅๆฐๆฎ้็จไบๅพฎ่ฐ
Under the overarching influence of a single, powerful force, China's open-source models filter out a large amount of content unfavorable to their designs during the pre-training stage, and then incorporate a significant amount of content defending their designs into the data constructed during the SFT stage. A corrective dataset is urgently needed for fine-tuning.

