惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

D
DataBreaches.Net
N
Netflix TechBlog - Medium
P
Proofpoint News Feed
D
Docker
J
Java Code Geeks
L
LangChain Blog
Microsoft Security Blog
Microsoft Security Blog
The GitHub Blog
The GitHub Blog
I
InfoQ
Stack Overflow Blog
Stack Overflow Blog
云风的 BLOG
云风的 BLOG
Engineering at Meta
Engineering at Meta
MongoDB | Blog
MongoDB | Blog
月光博客
月光博客
T
Tailwind CSS Blog
M
MIT News - Artificial intelligence
Blog — PlanetScale
Blog — PlanetScale
Google DeepMind News
Google DeepMind News
腾讯CDC
罗磊的独立博客
U
Unit 42
爱范儿
爱范儿
Vercel News
Vercel News
MyScale Blog
MyScale Blog

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL). mempalace/agent at agent · skorotkiewicz/mempalace
Customizing Karpathy's LLM wiki for fighting disease
kamens · 2026-04-28 · via Hacker News - Newest: "LLM"
personal disease wiki schematic

I now have two ‘personal disease wikis’ managed by AI. One for fighting type 1 diabetes (for myself) and one for multiple sclerosis (for a loved one). They stay up to date with the latest research, trials, technology, and even long-tail tips & tricks from patient communities online (which are often unproven, sometimes controversial, and consistently useful).

All of this, plus my personal medical records, gets combined into an organized wiki. The wiki comes with a personalized battle plan, mapping out options so I can be as proactive as possible.

This is the result of combining Karpathy’s LLM wiki idea with my hstack (healthstack) of skills and agents for anyone using AI to fight disease. It’s all open source at github.com/kamens/hstack. It’s a simple combination of ideas and markdown files that you can clone or steal.

Building my first personal disease wiki

init'ing my t1d llm wiki
Creating my T1D LLM wiki and ingesting personal records
t1d battleplan summary
Summary of the first ‘battle plan’ it created for me

Note the above unique combination of standard doctor advice, supplement suggestions typically avoided by clinicians, awareness of what’s happening with the latest technology, getting a jump on results from the latest T1D trials, practical glucose control tips anchored by my personal trends…

I’ve been working with doctors my whole life, and this amount of practical, proactive help wrapped up in one simple message would be rare to say the least. It’s like a mix of a great endocrinologist, a forward-thinking nutritionist, an empathetic patient community, a tech-forward diabetes expert, and an assistant who organizes my medical records.

The LLM wiki idea to build structured memory via simple markdown files means any non-technical user can open Obsidian and easily navigate around their personalized disease knowledge base.

Dumping a ton of messy personal medical record PDFs, clinicians’ notes, and lab results and having them cleanly organized into wiki structure is useful on its own, even before they’re personalized into a battle plan for you.

Why this specific context recipe works well for health

The results are interesting because they combine:

  1. Personal medical records
  2. Standard of care treatment options / best practices
  3. Bleeding-edge research papers, trials, new theories of disease
  4. Long-tail patient community threads, reddit posts, blog posts from people with your disease

Which means you wind up with something closer to what everybody wants when they interact with doctors:*

  • a battle-hardened expert in your disease
  • who is empathetic because they’ve internalized the nitty gritty of living with the disease
  • who’s reading the latest research and is great at mapping out new options
  • who’s not satisfied with just following standard of care and wants to take care of you the way they’d care for their own kid
  • who’s open-minded about the long-tail of community tips, tricks, and ideas that the medical industry is typically overly paternalistic about

Why Karpathy’s “LLM wiki” works well for health

  1. Simple human-readable markdown in a natural folder hierarchy is a friendly way to consume this sort of mess. You can install Obsidian on a family member’s phone, and they have an organized knowledge base about their loved one without needing to know about LLM wikis or how to talk to Claude Code or how to log in to an obscure portal.
  2. The ingest concept is simple for dealing with medical record hell - just download or screenshot the record, dump it into /raw, and run /hstack-wiki-ingest. No pretense that it’ll magically negotiate your healthcare portal data access rules. In the future an agent could help with this.
  3. The linting concept naturally becomes a refresh - /hstack-wiki-refresh updates with the latest research and patient community threads.
  4. Adding opinionated skills like those in hstack means you can talk to a diagnostic specialist, an R&D research specialist, a patient advocate, and more within the context of your personal disease wiki.

The LLM caveat

One of the best pieces of recruiting advice I got was: “Everybody spends all their time finding problems with candidates and then using interview after interview to pick at those problems. Instead, find what’s great and decide if the risk is acceptable.” Seems like a lot of people make this same mistake with LLMs.

Same story with hstack and this wiki idea. It’s a tool, not a doctor. It’ll say dumb stuff. It can’t examine you. It’ll also give you great options to think about. It’s up to you to decide what’s worthwhile / discuss with your doctors.

I’m used to ignoring 50% of what LLMs say. I just skim by the weak stuff. There are nuggets of gold to find.

More example nuggets of gold

refreshing t1d wiki
Running /hstack-wiki-refresh alerted me of the pending G6 Dexcom sensor discontinuation and my imminent need to upgrade
refreshing t1d wiki 2
The same refresh also made me aware of a new continuous monitoring concept I had no idea about
magnesium MS tip
My loved one with MS got this nuanced, useful tip that may help with an ongoing vitamin D absorption issue.

The magnesium glycinate tip for vitamin D absorption in an MS patient is interesting. It’s the sort of “evidence exists, but not so much that it’s a slam dunk for clinicians, but also it’s low risk, and the benefits could be nontrivial even if hard to prove” type advice that escapes the typical medical visit. It was spotted by hstack (A) recognizing a pattern in personal lab results, (B) being aware of the latest MS + Vitamin D research, and (C) being thoughtful about appropriate risks and personal lifestyle preferences.

Try it out

Claude can install this for you, instructions here: github.com/kamens/hstack


Much of this work came out of a few conversations with Ben Komalo, who was pushing me towards an “LLM wiki”-ish structure before Karpathy posted his tweets.

* Something about this being a totally indie, open-source, “use on your own or modify as you see fit” bundle of LLM skills makes it much easier to combine 1+2+3+4 than it would be if shipping a traditional product.

† Use it as a thinking partner, not a substitute for real medical advice. The authors are not liable for decisions made based on its output.