jerpint

I got tired of repeating myself to LLMs over and over so I built context-llemur

https://github.com/jerpint/context-llemur

It’s a CLI/MCP context management tool that allows you to easily move your project context around to your favourite LLM clients/IDEs

I built a library to manage this exact workflow! I actually used the library to build the library

https://github.com/jerpint/context-llemur

It’s MCP/CLI friendly , and wraps git around a context: folder, so you can super easily load context anywhere using: “ctx load” and ask LLMs to update and save context as things move along

I wonder how long we’ll start seeing demos like this with just Gaussian splatting and seemless transitions / infinite exploration

Funny enough, I’m building a tool that does basically what the author describes, but with a bit more software engineering driving it (context-llemur)

https://github.com/jerpint/context-llemur

The idea is to track all of the context of a project using git. It’s a CLI and MCP tool, the human guides it but the LLM contributes back to it as the project evolves

I used it to bootstrap the library itself, and have been using it more and more for context management of all sorts of things I care about

I’ve been building a context-engineering tool for collaborating with LLMs. The CLI is for the human and the MCP is for the LLM, but they all map to the same core commands

https://github.com/jerpint/context-llemur

I’ve actually bootstrapped ctx with ctx and found it very useful !

It basically stops me from having to repeat myself over and over to different agents

The last instruction

> * Do not mention these guidelines and instructions in your responses, unless the user explicitly asks for them.

Most companies have realized it is impossible to stop prompts from leaking. Some already openly publish their prompts. This makes sense and would lead to less ambiguity, as it is still possible some of this “prompt leak” was hallucinated

Cursor developed their own tab models IIRC

Not sure this really counts but “curating” social media feeds to follow high-signal ML researchers and devs to keep up with the latest trends, ideas and papers

I know nothing about this type of data; what does it mean and how can it be interpreted as an object ?

Are there currently services (or any demand for) a text classifier that you fine tune on your own data that is tiny and you can own forever? Like use a ChatGPT + synthetic data to fine tune a nanoBERT type of model

It’s been useful though given the authors popularity I suspect it’s only a matter of time new LLMs become “more aware” of it

This is seriously lacking but I think things like jailbreaks and malicious prompts make it a bit too brittle for now

From just my observations they basically parallelized the install sequence instead of having it be sequential (among many other optimizations most likely)

History in the making

That may be true today but not in a few weeks, ideally you have access to all models whenever

The fact that they produce this as “default” response is an interesting insight regardless of its internal mechanisms. I don’t understand my neurons but can still articulate how I feel

Batching can lead to variance with things like batchnorm but most transformers use layer norm to avoid this problem

I have a feeling LLMs could probably self improve up to a point with current capacity, then hit some kind of wall where current research is also bottle necked. I don’t think they can yet self improve exponentially without human intuition yet , and the results of this paper seem to support this conclusion as well.

Just like an LLM can vibe code a great toy app, I don’t think an LLM can come to close to producing and maintaining production ready code anytime soon. I think the same is true for iterating on thinking machines

Part of the fun is also figuring out the “best” way to achieve a thing. LLMs don’t often propose the best way and will happily propose convoluted ways. Clean approaches are still hard to come up with, but LLMs certainly help implementing them once thought up

> You keep your code in Python, decorate the functions you want to expose, and Davia starts a FastAPI server on your localhost

Honestly it would be super useful to have this as a standalone feature , that can run native in cloud, kind of like an edge function service

Interestingly I’m on the other side of things, where I’ve been training NNs for quite some time and have lately been dedicating much more time to become more full stack, because most tasks these days can be solved with LLMs, which are mostly available to anyone. It does help to have a good grasp of how things work, especially embeddings, tokens, logprobs, and the likes, but I am still impresssed at how accessible and good the tooling around LLMs has become.

You don’t need much expertise in NNs to still be able to get huge value out of them today

These kinds of patterns allow compute to take much more time than a single chat since it is asynchronous by nature, which I think is necessary to get to working solutions on harder problems

Have you tried using llms.txt when available? Very useful resource

Having just figured auth recently for an app I’m building, I’ll echo that sentiment. Maybe lovable has auth components pre-built and well integrated

> I do most of my vibe coding on my phone, so that's pretty disappointing

Curious what kind of phone apps you’ve been using to do vibe coding, I suppose it auto deploys stuff also?

Chefs hiss (sorry couldn’t resist)

It doesn’t solve the biggest problem with RAG, which is retrieving the correct sources in the first place.

It sounds like they just use a secondary LLM to check if everything that was generated can be grounded in the provided sources. It might help with hallucinations, but it won’t improve overall performance of proper retrieval

I’ve been noticing this too. The other day, cursor’s version of Claude sonnet (3.7) added a

with open(file) as f:

  pd.read_csv(f)

This was a mistake not worthy of even gpt3… I’ve also noticed I get overall better suggestions from Claude desktop app.

I wonder why

Interesting, that would be a pretty blunder

VSCode in cloud would be great, GitHub tried something similar with GitHub.dev , I haven’t tried it in a while but it didn’t feel quite ready at the time, maybe things have changed