Jacob Shea

Engineer, Founder, VC. Lettuce Grow Farmer, GP @StructureVC, Investor in Zero Waste, SaaS & Marketplaces & Good Humans. Mascot of Life. OBI LIFE.

[Jacob Shea @JacobShea]

Every damn day.

#GetItDone #Startups

[Jacob Shea @JacobShea]

@chamath I really found @karpathy's write up on the topic helpful, @lexfridman chimed in too

https://t.co/ZqkbopVsUH

thx Aaron & @karpathy

[Andrej Karpathy @karpathy]

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.

[Drew Breunig @dbreunig]

I get LOTS of questions about deploying DSPy programs, so @isaacbmiller1 and I built dspy-cli: a tool that serves DSPy programs as HTTP APIs with Docker config, OpenAPI specs, MCP support, and more.

Here's a quick intro:

[Jacob Shea @JacobShea]

Hey @bhalligan & @rabois, long time no chat. We just announced we're not raising a new fund, I'm ready to chat. Fantastic list, Brian.

1. Coder/builder: ✅ I wear glasses because of it.
2. Great taste in people: ✅ I run a group called Dorks with Attitude, collector of good people.
3. Contrarian: ✅ I would hope so!
4. Legit operating experience: ✅ I got scars.
5. Not an a-hole: ✅ I have daughters, how could I.

[Tomasz Tunguz @ttunguz]

"Context is the Defensible Moat: A generic LLM is a commodity; your business context is what makes your application defensible, because domain knowledge is the most valuable asset." - Yuanzheng Zhu, Instacart

Couldn't agree more.

Customizing AI systems with the internal knowledge of how a company works is the core of great AI software.

https://t.co/EpEx7xGfbA

[Jacob Shea @JacobShea]

@benln In! 🦾

[Jacob Shea @JacobShea]

@jacobhelberg @SergioGor Awesome Jacob, congrats! Long time coming.

[Jacob Shea @JacobShea]

@benln @angela_w Oh @angela_w! Congrats on the fund raise! Awesome news for you and the team.

https://t.co/09PsPY34jn

[Jacob Shea @JacobShea]

@benln hey neighbor @angela_w, let's get Ben Lang to host a Cursor meetup at Canopy :)

[Jacob Shea @JacobShea]

@angela_w @benln I agree! I'll send you some details on the community here at Canopy and it's surroundings

[Jacob Shea @JacobShea]

@RayFernando1337 @benln Please find me and say hello. By the printer :)

[Jacob Shea @JacobShea]

@benln Yes please!

[Jacob Shea @JacobShea]

@benln Helllloooo America.

[TBPN @tbpn]

We asked @ttunguz (Partner @theoryvc) about the value of data and feedback loops in AI.

"Initially, there's a lot of value in having access to proprietary data like YouTube or GitHub Copilot with initial access to GitHub. I think over time the more valuable and rare data asset is the feedback loop."

"This is also particularly true for distillation. Deepseek was hitting a bunch of APIs and copying really quickly."

"The larger the number of users you have and the greater the telemetry, the better the advantage."

[Jacob Shea @JacobShea]

After only three decades of reshaping the tech world in completely opposite directions, Linus Torvalds and Bill Gates have finally met.

Yes, that’s Dave Cutler the legendary engineer behind OpenVMS at Digital and Windows NT at Microsoft.

And Marc Russinovich is Microsoft Azure’s CTO and the legendary Windows engineer behind tools like Process Explorer. He’s the guy who knows how Windows works better than Windows.

[Jacob Shea @JacobShea]

This is what we need! Even therapists need help with their workload to help the masses. Well done Geoff and team.

[Anonymous Health @unwantedhabits]

🚨 We just went live on @NBCNewsNow @NBCNews!

Our CEO Geoff Nudd sat down with NBC to unveil the world’s first AI-Hybrid Care Model: designed to support people through addiction, depression, and anxiety.

Watch the full segment here:

[Jacob Shea @JacobShea]

@mufeedvh Looking forward to trying this out! Follow repo as well.

[Jacob Shea @JacobShea]

@GrindeOptions Not Unsupervised.
https://t.co/JmUv0Vzech

[Jacob Shea @JacobShea]

Coinbase partners with American Express to launch '@coinbase One Card' offering up to 4% Bitcoin back on all purchases. @pitdesi, what do you think? I love the HEX editor image, which reminds me of my 90s hobbies.

https://t.co/4EvNCUMzcQ

[Jacob Shea @JacobShea]

@anothercohen I’m back to … how I’ve missed …