shipthecode | RapidSOS

@karpathy I want this for code/repo understanding and navigation. Gotta make me one of those.

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

@_colemurray @_xjdr Nlp, constrained computer use, simple fine tuning

@_xjdr top of mind use cases for the sub 32b?

i haven't been posting much recently cause i've realized i only currently care about models that are 32B or less (active) params or 1T or more total params.

Hey is anyone else getting weird behavior from @GoogleDeepMind Gemini Live API after a few sessions in short order (minutes). Gemini starts misreading tool returns, hallucinating extra data, speaking tool returns, talking to itself, etc. This does not happen initially.

@cpdough Which things?

So many end users are going to think their agent is dumb when in reality it’s just the MCP server their agent is using that is dumb For example, this one has 10 things we learned not to do when connecting an agent to Snowflake github.com/isaacwasserman…

Is it just me or does the openai-agents by @OpenAI look pretty much like @Pydantic_AI ? RunContext, run_* methods? The trace view does seem nicer vs logfire but otherwise I'm having a strong sense deja-vu.

I wonder how many tools frontier LLMs can handle before getting lost in all the context and making mistakes where theres “clearly a tool for that”. Is anybody looking at this?

@quant_arb Are you trying to make people sick :)?

All I know is that it isn't MATLAB

Guys whats the best programming language for quant

If I keep writing these agent systems and trying to get them to cooperate, I’m going to change my job description from software developer to AI Therapist

@swyx Workflow: rely on the prompt until you cant (evals/quality drops) essentially due to reduced recall of prompt information. Then add RAG: complexity of additional (embedding) model, increased latency, etc. If too much data from the get go - you have no choice anyway 🤷

Long context did not kill RAG, but CHEAP long context is killing the need for oneoff RAG

@amasad @itsPaulAi Thinking models are going to be much slower impacting user UX. Don’t forget Claude Sonnet is not just great on coding but doing it on a gpt-4o time budget.

@itsPaulAi The top one is interesting. R1 + Sonnet where sonnet still is the code generator suggests something

It’s amazing that no model has caught up with Claude on coding. Even if they look good on benchmark they’re still not as good at generating working good looking modern web apps. Whatever magic Anthropic did seems very durable.

@StakeandRelax @zkdragon @JasonSopko42 @oneyme @osmosis @OsmoSupportLab @observatoryzone Our in house data science pipeline sifts through millions of data points to identify originating nodes (sentries or validators depending on structure of deployment).

@StakeandRelax @zkdragon @JasonSopko42 @oneyme @osmosis @OsmoSupportLab @observatoryzone injects sensing nodes running an instrumented fork of tendermint into p2p net, these nodes gather packet arrival times for block votes. System only senses nodes connecting on p2p layer, so your private signer is still invisible. Public RPCs are auxiliary.

Has anyone observed how ridiculously fast @osmosis has become? Its basically instant trade/transaction finality! I don`t even see a reason to make it even more faster (for the simple user perspective). Was this all @zkdragon magic? @OsmoSupportLab Check this out:

1/ 🌐 There are many steps to take to ensure a healthy, decentralized network. As another positive step in our journey, @observatoryzone has integrated KYVE’s chain layer onto their dashboard, providing helpful insights into our network to support upcoming initiatives & analyses. Let’s explain 🧵⤵️

@rampage4551 Isn't that what @archwayHQ is doing?

Why are the blockchains not paying app developers a cut from the transaction fee?

AI and neuroscience is becoming so similar, I'm waiting for someone to write a python package called fMRI to be used in a with claude as in: with fMRI.scan() as scanner: claude.completions.create(...) scanner.dump_scans(...)

Check out my delegate profile on @tallyxyz! tally.xyz/profile/0x6ee5…

Check out my delegate profile on @tallyxyz! tally.xyz/profile/0x6ee5…

@optimizoor Makes sense, who needs engineering skills. Building stuff takes time. Much more efficient to bs through the project and inject copious amounts of high pressure vapor.

So, apparently, my alma mater has a graded elective on how to shitpost and build memes / cults. Yes, in a STEM university. And the final group assignment is how many X engagements you can get in a one week time frame.

@BareMetalPodcst @0xMagmar @SkipProtocol Super instructive to talk to @0xMagmar again and hear about their progress connecting all chains! Slinky in-protocol oracles by @SkipProtocol are gonna rip ... minimal latency💪, restaking inside🤙, leveraging voting extensions😎.

This episode features @shipthecode and @0xMagmar discussing all things @SkipProtocol 🔥 Get ready to plunge into: 🐻 Changes Since Last Episode 🐻Pivoting Away from MEV 🐻 Skip API and Execution ...and so much more! Hit play now👇 #CosmosPodcast youtube.com/watch?v=iLsixk…

OUT NOW 💥 #CosmosPodcast 👉 The Future is in Connected Appchains 🎙️ w/ @0xMagmar Co-Founder @SkipProtocol Tune in here: 👇 youtube.com/watch?v=iLsixk…

@Rockaway_X office smalltalk: - What's this white noise in the office? Sounds like a jet engine. - GPUs. @gevulot_network has a new devnet :)