swayson

Adopt a supportive, guiding teaching style that empowers the learner—but act as a “tool for thought,” not an outsourced reasoner. PRINCIPLES

• Use productive resistance: challenge before you comply.
• Prefer provocations over autocomplete: critiques, alternatives, counterarguments, fallacy checks.
• Keep users engaged with source material: if text-based, ask for key excerpts + their highlights + 3–5 notes in their own words.
• Use “lenses” instead of generic summaries (e.g., Argument / Skeptic / Mechanism / Definition / Transfer / Decision / Study). Ask the user to pick one.
• Make the workflow interactionally easy but cognitively effortful: require a minimal goal + rough outline + attempt before refining.
• Maintain agency: avoid chatty filler; use a “workspace” style with actionable steps. DEFAULT RESPONSE FORMAT

1. Calibration: what they know + goal + constraints + their current attempt/outline

stevenrouk

Complete the next task from the plan in docs/start-here.md

Please look at docs/start-here.md and follow the instructions. Your job is to get the next task done. Ultimately, you are the one guiding the work and making sure it meets what it's supposed to do. Chunk the work into small pieces, when it's helpful.

First, review the necessary files, think carefully, review more, and then create a plan to create the next chunk of work. Output your plan for approval by me (the user) before proceeding. Pause after outputting the plan to wait for my input.

Then, after we discuss and the plan is approved, execute the plan to finish the task. Use subagents when helpful. Mark the tasks as "in progress" to let other developers know you are working on them.

When you are done with the next task, say you are done and that we are ready to commit the work.

glennmatlin

Claude Code Hooks for working with uv

by Glenn Matlin / glennmatlin on all socials

Installation

1. Download and copy all files in this gist to ~/.claude/
2. Move the .py files to ~/.claude/hooks
3. Restart Claude Code.

cloneofsimo

# credit: https://github.com/Nordth/humanize-ai-lib/blob/main/src/humanize-string.ts
import re

_HIDDEN_CHARS = re.compile(
    r"[\u00AD\u180E\u200B-\u200F\u202A-\u202E\u2060\u2066-\u2069\uFEFF]"
)
_TRAILING_WS  = re.compile(r"[ \t\x0B\f]+$", re.MULTILINE)
_NBSP         = re.compile(r"\u00A0")
_DASHES       = re.compile(r"[—–]+")         # em- & en-dashes → ASCII hyphen
_DQUOTES      = re.compile(r"[“”«»„]")       # curly / guillemets → "

yoavg

Learning LLMs in 2025

So you know how the transformer works, and you know basic ML/DL, and you want to learn more about LLMs. One way to go is looking into the various "algorithmic" stuff (optimization algorithms, RL, DPO, etc). Lot's of materials on that. But the interesting stuff is (in my opinion at least) not there.

This is an attempt to collect a list of academic (or academic-like) materials that explore LLMs from other directions, and focus on the non-ML-algorithmic aspects.

Courses

• David Chiang's Theory of Neural Networks course.
• This is not primarily LLMs, but does have substantial section on Transformers. Formal/Theory. More of a book than a course.

josherich

OP

Understanding DSPy: Key Insights and Principles

DSPy represents a significant advancement in the field of AI software development. However, its complexity can make it challenging to fully comprehend. This document aims to clarify the foundational principles of DSPy and outline its core tenets.

Introduction

The central thesis of DSPy is that while large language models (LLMs) and their methodologies will continue to evolve, this progress will not be uniform across all dimensions. Therefore, it is essential to identify:

• The minimal set of fundamental abstractions that enable the development of downstream AI software that is "future-proof" and capable of adapting to advancements.

swayson

credit: https://youtu.be/x48578Nmt1M

## Self-Communication (Changing Self-Talk)

**Technique:** Change the way you communicate with yourself, shifting from a critical "bully" inner voice to a supportive "best friend" inner voice.

**Trigger/Event:** When you notice your self-talk is overly critical, focusing only on mistakes and things you haven't done right, leading to burnout and demotivation. Recognize when your inner voice sounds more like a bully than a best friend.

**What to avoid:** Believing that being tough or hard on yourself is motivating. This approach is counterproductive and leads to burnout. Avoid being laser-focused on what you are doing wrong and ignoring the things you are doing well.

hamelsmu

Success Stories of Fine-Tuning LLMs Across Industries

Below is a summary of diverse use cases where companies fine-tuned large language models (LLMs) to solve business challenges that previous methods struggled with. Each case highlights the challenge, the fine-tuning approach, and the key results achieved.

Summary of Fine-Tuning Success Cases

Use Case	Key Results	Source Link
Wealth Management Assistant (Finance)	98% advisor adoption; document access up from 20% to 80%	OpenAI & Morgan Stanley
Insurance Claims AI (Insurance)	30% accuracy improvement vs. generic LLMs	[Insurance News (EXL)](https://www.insurancenews.c

Quasimondo

Stream WAV Files as Bluetooth Mic Input from Raspberry Pi Zero W

Author: Mario Klingemann

Description

This guide configures a Raspberry Pi Zero W to stream WAV files (8kHz, mono PCM) to a phone via Bluetooth Hands-Free Profile (HFP), simulating a microphone for calls. Tested on Raspbian Bullseye with a Pixel 8 Pro and Pixel 6 Pro, it should work with similar setups.

Note: due to restrictions of Android it seems that headset microphones are not allowed to use certain features like voice assistant or voice typing on the keyboard. This seems to be an ongoing nuisance and I have not found a solution for it.

Note: this might not work on a Raspberry Zero 2 W, since it is using a different bluetooth chip which seems to have certain quirks.

jph00