Joon Choo

@karpathy What I’ve found to be helpful to producing these wikis is NoteBookLM. it can ingest 200MB of files, articles, YouTube links per notebook and produce markdown files with shocking levels of granularity and coverage without hallucination. A “lower tech” way to produce these en masse

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.

@orrdavid Happened on this tweet while revising my position - was wondering if people were still short since IKs hit 7-10%+ RLM in Naperville and management want to retrofit to 25-50% stores, and 18-20% unit growth until 2030 seems pretty steep to me

I've been short Sweetgreen $SG but I'm tempted to quit. A friend *really* loves it, says the place is very busy in LA and he thinks they can raise prices decently. If they can, and if they can expand, it could work very well long term. Any opinions?

7/ If you then add the cost of contractors, architects and subcontractors (according to HNGRY) is around $140K cash per store. What are people’s thoughts on the economics here?

3/ Had chat with hyphen/$CMG investors who say it contributes to the notion of significantly overstated throughput – higher throughput/capacity unlocks flywheel of lower wait times → increased demand during peak hours → higher utilization of fixed labor costs.

Been looking into $SG and was wondering what people think. 1/ I understand that there seems to be a lot of interest in EBITDA margin expansion on a cost covered basis with IKs but what about the opportunity cost and cash conversion from opening new stores?

6/ Add retrofit costs and the lost profit of a store closure to the IK’s $500K product cost. If an NYC store does $3.5M in sales, the opportunity cost of the retrofit is ~$53K per month ($3.5M/12 x 18%).

6) Quoted from HNGRY "Take the new Penn Plaza location, (assuming management’s 10% margin boost and $450K-550K reported CAPEX)" that means that infinite kitchen payback at the first two $2.6mm AUV suburban locations is about two years. How about retrofitting urban stores in NYC?

5/ thoughts on street models? SG priced to near perfection – street DCFs incorporate 2/3rds of sweetgreen’s with IKs after hitting 1k locations with $3.8mm AUVs by 2036. Limited proof of concept and high opportunity cost of retrofitting. Breaking down numbers next thread.

4/ Read a HNGRY article cannel check says 1 employee processes ~1 bowl/30 seconds --> 120 bowls per hour within the Infinite Kitchen (so 240/hour for 2 employees). Around 31% lower than Spyce’s original statement of 350 bowls per hour in its own retail locations

@techinvestoor @NEWBUHO 2/ Seems that $CMG with Hyphen and 1.5 years proof of concept with $SG + IKs has ended debate about automation, most street models bake in 7-10%+ margins. Hearing that most investors care more about the revenue story (so SSS + unit growth) was wondering what your thoughts were

@NEWBUHO The undervaluation is absurd when you think of the potential for them to expand nationally and internationally and when you factor in 10%+ profit margins from robotic kitchens. The next $CMG? I think so for $SG.

$SG $50 up next then $100 then $200 Dramatically undervalued relative to opportunity in front of them to open hundreds of Infinite Kitchens with 10%+ higher gross margins.

@techinvestoor @NEWBUHO 1/ I understand that there seems to be a lot of interest in EBITDA margin expansion on a cost covered basis but what about the opportunity cost and cash conversion from opening new stores?