Behnam Moh ibehnam

Because I work on AI, people sometimes email/DM me and ask for a meeting to discuss their ideas about a paper/project. I already have a lot on my plate so collaborating with them is not an option. Still, I often read their messages and agree to chat, despite having a tight schedule. Sometimes the conversation goes like this:

Them: Can we talk sometime this week?
Me: Sure, how does Friday 2pm sound?
Them: Can we do it earlier, like 11am?
Me: I can’t, there’s a seminar I have to go to.
Them: Okay, 2pm actually works for me.

At this point, they’ve already shown their disrespect for my time: 2pm would work for them but they prioritized their convenience over mine. But it gets worse: these people typically then ask me to send them a calendar invite or Zoom link!

Latency numbers every programmer should know

L1 cache reference ......................... 0.5 ns
Branch mispredict ............................ 5 ns
L2 cache reference ........................... 7 ns
Mutex lock/unlock ........................... 25 ns
Main memory reference ...................... 100 ns             
Compress 1K bytes with Zippy ............. 3,000 ns  =   3 µs
Send 2K bytes over 1 Gbps network ....... 20,000 ns  =  20 µs
SSD random read ........................ 150,000 ns  = 150 µs

Read 1 MB sequentially from memory ..... 250,000 ns = 250 µs

Extracting financial disclosure reports and police blotter narratives using OpenAI's Structured Output

tl;dr this demo shows how to call OpenAI's gpt-4o-mini model, provide it with URL of a screenshot of a document, and extract data that follows a schema you define. The results are pretty solid even with little effort in defining the data — and no effort doing data prep. OpenAI's API could be a cost-efficient tool for large scale data gathering projects involving public documents.

OpenAI announced Structured Outputs for its API, a feature that allows users to specify the fields and schema of extracted data, and guarantees that the JSON output will follow that specification.

For example, given a Congressional financial disclosure report, with assets defined in a table like this:

Purpose

Bootstrap knowledge of LLMs ASAP. With a bias/focus to GPT.

Avoid being a link dump. Try to provide only valuable well tuned information.

Prelude

Neural network links before starting with transformers.

Karabiner layouts for symbols and navigation

Gavin Sinclair, January 2022

Introduction

I use Karabiner (configured with Gosu) to make advanced key mappings on my Apple computer. Karabiner allows you to create “layers”, perhaps simulating those on a programmable mechanical keyboard. I make good use of these layers to give me easy access (home-row or nearby) to all symbols and navigational controls, and even a numpad.

The motivation is to keep hand movement to a minimum. Decades of coding on standard keyboards has unfortunately left me with hand and wrist pain. I will soon enough own a small split keyboard which will force me to use layers to access symbols etc., so this Karabiner solution, which has evolved over months, is a training run for that.

	#cloud-config

	# Enable automatic package updates and upgrades during cloud-init execution
	package_update: true
	package_upgrade: true

	packages:
	# Security and Hardening
	- ufw
	- fail2ban


	Jeff has two brothers and each of his brothers has three sisters and each of
	the sisters has four step brothers. How many step brothers does each brother
	have?

	<think> Okay, so I've got this problem: Jeff has two brothers, and each of his
	brothers has three sisters. Then it says that each of the sisters has four step
	brothers. The question is asking how many step brothers each brother has.

	Hmm, let me start by breaking down the information piece by piece. First, Jeff

	Question: Jeff has two brothers and each of his brothers has three sisters
	and each of the sisters has four step brothers. How many step brothers does
	each brother have?


	Model's Response:


	<think> okay, let's try to figure out this problem step by step. so jeff has
	two brothers. that means there are three boys in total: jeff and his two

	font-family = BerkeleyMono Nerd Font
	#font-family = Iosevka Nerd Font
	# font-family = SFMono Nerd Font
	font-size = 20
	theme = GruvboxDarkHard
	shell-integration-features = no-cursor,sudo,no-title
	cursor-style = block
	adjust-cell-height = 35%
	# background-opacity = 0.96

	{
	;; :!Ca is keycode :a and prefix a with !C

	;; C \| left_command
	;; T \| left_control
	;; O \| left_option
	;; S \| left_shift
	;; F \| fn
	;; Q \| right_command
	;; W \| right_control