just some web dev
@randocodemonkey.bsky.social
what it says on the tin.
research notes & advice soliciting
research notes & advice soliciting
* 99/1 split
November 18, 2025 at 4:10 PM
* 99/1 split
this gets at my worry about over-trust on AI as like an industry.
the human mind craves flow. interruption takes it out of that. there's a strong bias toward either (or both?) "let it cook" or "I'll do it myself". a 99/1 does not jibe
the human mind craves flow. interruption takes it out of that. there's a strong bias toward either (or both?) "let it cook" or "I'll do it myself". a 99/1 does not jibe
November 18, 2025 at 4:10 PM
this gets at my worry about over-trust on AI as like an industry.
the human mind craves flow. interruption takes it out of that. there's a strong bias toward either (or both?) "let it cook" or "I'll do it myself". a 99/1 does not jibe
the human mind craves flow. interruption takes it out of that. there's a strong bias toward either (or both?) "let it cook" or "I'll do it myself". a 99/1 does not jibe
answers, not good ones seems to be the norm! which I guess is fine if you're willing to spend time with it and massage the results. trust but verify, etc.
November 18, 2025 at 12:28 AM
answers, not good ones seems to be the norm! which I guess is fine if you're willing to spend time with it and massage the results. trust but verify, etc.
also the info on Ollama in context of codebases is olllld. my sniff-test isn't well-calibrated for AI, but that use case seems to be hitting that "old and smelly" threshold
November 18, 2025 at 12:06 AM
also the info on Ollama in context of codebases is olllld. my sniff-test isn't well-calibrated for AI, but that use case seems to be hitting that "old and smelly" threshold
ahh, so is non-code data a better use case for vector DB / embeddings? it seems like some folks have integrated it into their workflow, it's just really weird how low the volume/quality of examples
November 18, 2025 at 12:04 AM
ahh, so is non-code data a better use case for vector DB / embeddings? it seems like some folks have integrated it into their workflow, it's just really weird how low the volume/quality of examples
Ollama seems cool. running the model locally w/ latency and privacy benefits appeals. for RAG it seems like you have to create a vector DB and generate embeddings to populate it, though, and examples are scarce?
honestly tho, the article I found might just be wack. idk
medium.com/@farissyaria...
honestly tho, the article I found might just be wack. idk
medium.com/@farissyaria...
Ask Your Codebase Anything Using Ollama, Embeddings, and RAG
In this tutorial, we are going to create a simple Retrieval-Augmented Generation (RAG) system that allows you to query your code repository…
medium.com
November 17, 2025 at 11:24 PM
Ollama seems cool. running the model locally w/ latency and privacy benefits appeals. for RAG it seems like you have to create a vector DB and generate embeddings to populate it, though, and examples are scarce?
honestly tho, the article I found might just be wack. idk
medium.com/@farissyaria...
honestly tho, the article I found might just be wack. idk
medium.com/@farissyaria...
this thread has some good thoughts too
feeling a bit "done" with claude code, just tired of it constantly trying to half-ass things. gonna give codex a try?
November 17, 2025 at 11:20 PM
this thread has some good thoughts too
I'm at a VS code shop but looking at what the options are. Zed seems to have some clout atm
so you're saying it has out of the box support for git, vim and c++, and it's foss? ok fine I'll try zed
November 17, 2025 at 11:16 PM
I'm at a VS code shop but looking at what the options are. Zed seems to have some clout atm