Laurian Gridinoc Laurian

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

macOS Internals

Understand your Mac and iPhone more deeply by tracing the evolution of Mac OS X from prelease to Swift. John Siracusa delivers the details.

Starting Points

How to use this gist

You've got two main options:

ChatGPT Resources

Context

ChatGPT appeared like an explosion on all my social media timelines in early December 2022. While I keep up with machine learning as an industry, I wasn't focused so much on this particular corner, and all the screenshots seemed like they came out of nowhere. What was this model? How did the chat prompting work? What was the context of OpenAI doing this work and collecting my prompts for training data?

I decided to do a quick investigation. Here's all the information I've found so far. I'm aggregating and synthesizing it as I go, so it's currently changing pretty frequently.

Model Architecture

Circumventing Deep Packet Inspection with Socat and rot13

I have a Linux virtual machine inside a customer's private network. For security, this VM is reachable only via VPN + Citrix + Windows + a Windows SSH client (eg PuTTY). I am tasked to ensure this Citrix design is secure, and users can not access their Linux VM's or other resources on the internal private network in any way outside of using Citrix.

The VM can access the internet. This task should be easy. The VM's internet gateway allows it to connect anywhere on the internet to TCP ports 80, 443, and 8090 only. Connecting to an internet bastion box on one of these ports works and I can send and receive clear text data using netcat. I plan to use good old SSH, listening on tcp/8090 on the bastion, with a reverse port forward configured to expose sshd on the VM to the public, to show their Citrix gateway can be circumvented.

Rejected by Deep Packet Inspection

I hit an immediate snag. The moment I try to establish an SSH or SSL connection over o

Hacking the Rectangular Starlink Dishy Cable

These are are some notes I put together on butchering the rectangular dishy cable.

FOLLOW THESE GUIDELINES AT YOUR OWN RISK. I TAKE NO RESPONSIBILITY FOR ANY DAMAGE OR INJURY YOU SUSTAIN FROM FOLLOWING OR NOT FOLLOWING THESE GUIDELINES.

	base_model: meta-llama/Meta-Llama-3-70B
	model_type: LlamaForCausalLM
	tokenizer_type: AutoTokenizer

	load_in_8bit: false
	load_in_4bit: true
	strict: false

	datasets:
	- path: /home/migel/ai_datasets/tess-v1.5b-chatml.jsonl

	import llama_cpp
	import re
	import json

	# Model configuration
	# tested with mistral, llama2, llama3, and phi3
	model_path = "/path/to/model"

	base_llm = llama_cpp.Llama(model_path, seed=42, n_gpu_layers=-1, n_ctx=4096, verbose=False, temperature=0.0)

	// HypothesisEmbed.vue
	// This is the component that mounts on the page that has the annotations
	// It's important to destroy the client when componenent is unmounted
	// because if the component remounts, it creates duplicate annotations in the DOM

	<template>
	<div
	id="HypothesisEmbed"
	class="HypothesisEmbed"
	/>

	import z from "zod";

	export async function safeFetch<T>(
	schema: z.Schema<T>,
	input: RequestInfo,
	init?: RequestInit
	): Promise<T> {
	const response = await fetch(input, init);

	if (!response.ok) {

	// I'm tired of extensions that automatically:
	// - show welcome pages / walkthroughs
	// - show release notes
	// - send telemetry
	// - recommend things
	//
	// This disables all of that stuff.
	// If you have more config, leave a comment so I can add it!!

	{