Machupo

Synthetic biologist into design and process optimization. Sliding sideways into NN dev and human-machine cognitive teaming. We design & effect our own future.

[Machupo @Machupo1]

@DJSnM @blueorigin Yeety McYeetface?

[Machupo @Machupo1]

@Teslarati Still on .3.2, but it seems to have figured out my preference for backing into parking spots (does it just about every time); still working on backing into the garage after homelink auto-activation.

[Machupo @Machupo1]

@Teslarati I must be in the penalty box update segment; still running 14.2.2.5 :/

[Machupo @Machupo1]

@Teslarati I am going to lump incorrect speed limits in with nav error disengagements

[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.

[Machupo @Machupo1]

@Teslarati Homelink activation from app; no garage door opener outside my house, and carrying my phone anyway. Very helpful when doing yard work and someone closes the garage door.

[Machupo @Machupo1]

@Teslarati This is also why there is reinforcement learning / fine tuning, but I don’t know what the FSD team focus is for that (probably more important stuff than my gripes, lol)

[Machupo @Machupo1]

@Teslarati A problem with training data sourced/synthesized from humans is it misses easy robot-precision wins = why profiles don’t do granular mph max, why FSD weaves instead of exact lane centering, and the post-merge slowdown. Some behaviors are not colinear and should not be integrated.

[Machupo @Machupo1]

@Teslarati FSD needs a back into a spot option.

[Machupo @Machupo1]

@ray4tesla If only Tesla had made a garage door opener module that was linked to your home and then included that in the FSD stack… seems like an obvious missed opportunity.

[Machupo @Machupo1]

Few days with FSD 14.1.4 and it has learned me (or I have learned it). No disengagements in 40min of fast-paced rush hour and only one wheel-torque nag. MM for highway, STD for most roads, and Sloth for the neighborhood. Now to sort homelink integration and garage waypointing.

[Machupo @Machupo1]

1st FSD 14.1.4 ride - four takeovers. Two due to incorrect speed limit detection, one to stop an evasive maneuver that was unnecessary, and one in MM mode when it hesitated and I almost missed my interstate exit. Generally very impressed with the update - feels much more natural.

[Machupo @Machupo1]

@elonmusk KSR did it with 100, Starship fits that envelope (at least to an orbital construction yard). Any plan for larger multiple-SH transit vehicles or just planning to overcome radiation risk with numbers?

[Machupo @Machupo1]

@elonmusk TTS would be nice. Would also like gestural and natural language feedback for FSD reinforcement learning (faces are high info density, per snow crash).

[Machupo @Machupo1]

@IMPAarif @Tesla It doesn't care, hits them at full speed. The problem is not just potholes, it is a lack of IRL (inverse reinforcement learning) feedback handling from camera or natural language input. If the NN could correlate my swearing with its hitting potholes, maybe it would learn faster.

[Machupo @Machupo1]

@RubenClaeys Mahna Mahna from the Muppets https://t.co/DfgJ5KpCN3

[Machupo @Machupo1]

@wholemars Installing now -- looking forward to autopark from FSD! Wonder how it will work at superchargers, if it will pick an open stall (13.2.7 would dump me somewhere in the parking lot of a supercharger).

[Machupo @Machupo1]

@elonmusk 20 million centenarians?!?! Woah.

[Machupo @Machupo1]

@Teslarati As many have already said, end-to-end FSD makes this largely irrelevant. But for the purpose of completeness, buttons and yoke in my M3P. Even in roundabouts, finding the appropriate signal to exit isn't rocket engineering.

[Machupo @Machupo1]

@Teslarati The road barrier does have the shape of the Cybertruck's potential mate. Obviously love at first sight.