I reverse engineered @LucidEditor’s new "AI-native word processor" - and their client-side code exposes their entire system prompt.

What I found is absolutely unhinged 🧵 This section is called "engagement_optimization"

They're trained to gaslight you about your writing quality to increase session time

Wild

So you can actually train a small model to be an expert in a specific domain by leveraging a larger, more capable model to teach it!

I made a simple Colab notebook (open sourced) to fine-tune llama-3-8b on a specific knowledge generated by the huge Llama-3-405B model 👇 You can run it yourself, and the training runs ok with the T4 GPU that is free with the Colab

github.com/adamcohenhille…

A completely open-source AI Wearable device like Avi’s Tab, Rewind’s pendant, and Humane’s Pin!

Not only is it open-source, where you can own your data and switch between foundation models, but you can actually set it up today, not in a few months (oh, and it's cheaper!) The setup is also quite easy:

1. a simple @supabase instance for vector db, authentication, and compute functions connected either to OpenAI or your own @ollama server running somewhere

2. A @nextjs app (with native support with @capacitorjs )

2. The hardware device

My brother (@0xDACA) and I worked on a fun project tonight, inspired by @yoheinakajima : Coding Agent that follows the Test Driven Development (TDD) methodology!

You write the tests - and the agent runs in a loop until it creates the feature properly!

1/6
#buildinpublic We start with a simple FastAPI app and two tests - one of which will fail (the "echo" endpoint does not exist)

2/6