Arun Sharma
@arundsharma
Databases | Encrypted Chat | Ranked Choice | Single Dad
x.com/lateinteractio… This. Giving a model access to an interpreter, web search and a few embedded storage engines is the future. What should we call this system?
We might be departing the stage where frontier foundation models are "models" under any sensible definition. They are now going to all become squarely compound systems, as far as I can tell? Reasoning models are the last ambiguous stage.
People using @PerplexityComet: are you able to fetch your twitter feed locally and do some filtering/reranking yourself? Would love to hear from OpenAI/Anthropic agent users too.
Is there a github with the configs used in this video that work out of the box on Ubuntu? I'm an arch user too, but prefer to run it inside a container on ubuntu. omarchy.org
leoniemonigatti.com/blog/what_i_le… Well written post by @helloiamleonie. Someone should write an addendum about graph based retrieval as something that complements keyword, vector and db index based retrieval.
github.com/adsharma/ask-m… If you're looking for an Electron desktop app to chat with your LLMs.
Search Ranking people are the inventors of things that became generative AI. How are their jobs safe from AI? Specifically does this pre-2022 view still hold or should we give a few MCP storage tools to AI and it figures things out? x.com/jobergum/statu…
Quality retrieval has never come from general-purpose storage or SQL databases. Decades of search history (web, enterprise, recommendations) show that good retrieval systems are specialized, with purpose-built ranking. Storing a vector doesn’t cut it. The category ‘vector…