I recently needed to setup a new Python project and in the course of doing so, learned about uv¹ , “an extremely fast Python package and project manager”. uv is great. It is easy to install (just a single executable!), and it takes care of all the essential elements of a Python project. Unfortunately, I immediately came across one major problem–uv uses the wrong default for project layouts.

This is the default layout of a project created by uv:

Farewell to Google Search

Let’s say you were in the market for a new laptop. What would you do? You certainly would not go on Google and search for “A good laptop under $1000”. That’d be madness. It used to be that Google could be a trusted source of information. These days, however, the top results on Google are mostly advertisements.

Generative AI 2024 Retrospective

2024 witnessed a parade of increasingly more powerful AI model releases, culminating in OpenAI’s groundbreaking “o3”. While I’m certain these advancements really are significant, I don’t think they have translated to noticeable improvements for the average user.

What people are paying attention to, though, are the really amazing product features that are coming out of Anthropic and Google. I’ve heard a lot about people building software Claude’s Artifacts. And Google’s Deep Research product, released a few days ago, has just completely changed search.

This is a review of Open WebUI, an extensible and user-friendly self-hosted WebUI for LLMs.

I recently decided to run my own UI layer for LLMs, as I have some exciting ideas. For those who are not familiar with chatbots, their architecture basically looks like this:

While I’m pretty good at customizing the middleware, I’m not as skilled at customizing the UI layer.

I tried writing my own UI at first, but I quickly gave up when I realized it was beyond my skills as a frontend developer. The next thing I did was look into open-source solutions. There were a lot of choices, but I narrowed them down to these three:

OpenAI took the world by storm in 2023 with the relase of ChatGPT. Other companies quickly followed suit, introducing their own competitors to ChatGPT. In this update, I want to write down my thoughts on some of the biggest players in the field of generative AI.

When I use the term “generative AI,” I am referring to a computer program that can generate text, images, or other types of content. These programs can understand natural language and follow complex instructions. One notable example of a generative AI is OpenAI’s “GPT-4,” commonly known as ChatGPT. Another well-known generative AI is Midjourney, which specializes in generating images from text descriptions. In this update, I will focus on AI with text output, as that is the area I have the most experience with.

Once you know a programming language well, the process for learning a new language is not very hard. It is just time consuming. You need to read the documentation for basic syntax and flow control, get familiar with its idioms, memorize core parts of the standard libraries, and learn its tool chains. What can we do to speed up the learning process? One thing we can do is provide great examples in the documentation. Can we do better? What if you have working examples for every problem you encountered? What if you can describe your intents in a familiar language and see how it should look in a new language? As it turns out, GPT-3 is really good at this task.

Writing code with ChatGPT.

In the aftermath of ChatGPT’s debut, the Internet is abuzz with GPT’s ability to write code. I used it to write code yesterday, and this is what I thought.

I wanted to create a command line program to interact with APIs from OpenAI and AI21. The design is a single program with subcommands for each company’s APIs:

Usage: /bin/foo_to_be_named <command>

A tool to interact with various LLM models from the CLI

Flags:
  -h, --help                                     Show context-sensitive help.
      --conf-file="llmcli.conf"    file with api keys, created with http://localhost:8080/v1/p/tools

Commands:
  openai create-completion
  openai create-edit
  ai21 complete

Run "/bin/foo_to_be_named <command> --help" for more information on a command.

I like to use Kong as the framework for command line parsing and I wanted to know how to use its subcommand features. Being too lazy to read the documentation, I asked ChatGPT to write me a short program using sub-commands. This is the example it came back with:

I had two weeks off at the end of 2022. Telesign closed its operations on the last week of 2022 to give employees well-deserved time off for a year of hardwork and I took another week off in addition to that. Like many technologists, I became captivated by OpenAI’s release of ChatGPT in 2022, and I spent a lot of the last two weeks exploring what OpenAI has to offer.

ChatGPT is a chatbot developed by OpenAI. It has been widely recognized for its impressive capabilities, such as imitating human writing, transforming plain English into code, and making glaringly stupid mistakes. The underlying technology of ChatGPT is a machine learning model known as GPT-3. This model is designed to predict how a human might continue a previous piece of text. For example, given what I’ve written so far, GPT-3 predicted what the rest of this blog post will look like:

Author: Rachel Simmons

Notes

• It’s not always about your child. She must learn what we have and do not have control over
  • Examples: Racism, sexism, stereotypes
  • Be aware of such injustices. Learn to manuver around them
• Model self-regulation
  • Example: Not panic when your child gets poor grades
• “I wrote ‘be more confident’ in my previous …. I have a change of heart.” - Simmons

Principles

• Model serenity in the face of uncertainty
• Calibrate parental expectations
• Manage your own parental anxieties

Tips

• Identify facts. Be deliberate in parenting
• Do not make colleage the goal
• Do not “help her” lose weight; Call out unnaturally thiness in advertising and on-screen
• Tone down the Girl Power talk
• Illustrate that failure is part of the journey with Failure Resume
• Set realistic goals, e.g., take small risks before taking big ones
• While counseling both emphasize and encourage making decisions
• Practice self-compassion
• Mindfulness; observing what you’re feeling withou judgement
• Self-kindess; saying something kind to yourself
• Thinking about others who share your experience
• Identify your self-critic as a separate entity
• Even grown-up children have tantrums; react accordingly
• Teach not everything can be fixed

Assertions

• Role models have made the Supergirl a baseline – creating unrealistic expectations
• Social media is damaging to a girl’s mental health
• Making college the most important goal is harmful and disincentivize courage taking
• Females care more about pleasing others, seeking feedback, performing and looking good
• A full schedule harms by confusing plan for purpose
• We send the message that a failure to have a career goal by high school is a failure in life
• Even in the 21st century, a fat female body is still expected to meet an unrealistic expectation of beauty
• A lack of confidence is not her fault
• Confidence can be learned - by hearing no repeatedly
• Defensive perssimism (“I’m going to fail”) is commonly practiced by girls
• Girls are more prone to rumination, which is largely cultural
• Girls struggle at finding kind words for themselves
• Our culture expects impossible standards of girls

Claims

• Academic performance does not suffer when parents do not emphasis acheivements
• For girls, there epidemic of poor mental health and lack of confidence
• Children who gets rewarded for an activity los intrinsic motivation for that activity; girls especially vulnerable
• Researches attempting to linking social media to well-being have been inconclusive
• Girls are prone to social comparison and feeback seeking online
• Body is the number two worry for young girls
• Affulent girls are more vulnerable to body image issues
• There is a confidence gap between males and females
• Girls are at greater risk for emotional problems
• Girls practice co-rumination which undermines their emotional well being
• Girls report more shame then boys
• Black girls are known to lose the least self-esteem during adolescence

Unanswered questions

• What is the right level of expectatio for any given family?
• Why extrinsic rewards harm females more?
• Is it because males are, as an unwritten rule, evaluated via the single metric of wealth? i.e., society and parents align on a single metric
• How to address the paradox of empowerment and self-blame?
• Where is the fine line between thinking hard vs rumination?
• Where is the fine line between self ciriticism and lack of confidence?
• Where is the fine line between asking for help and learned helplessness?
• When to preservere vs when to give up?
• Is this a form of Optimal Stopping Problem? N-Armed Bandit Problem?
• When to accept what one cannot control, when to have courage to change things

bind

Sometimes a synonym for “map”

conflate

combine

disjoint

separate, e.g., odd numbers and even numbers are disjoint

disjunction

inclusive or – if one of the inputs is True

extrinsic

not intrinsic

federated

Top-down delegation of responsibilities; Has a single point of failure at the top.

PID controller

control loop feedback mechanism

e.g., curise control

99%ile

Abbreviation of percentile

EBNF

Extended Backus-Naur Form

Useful for defining the syntax of a programming language

EBNF terminal

a token/word/chunk in EBNF

Alpha

Αα

Beta

Ββ

Gamma

Γγ

Delta

Δδ

Commonly denotes ‘difference’

Epsilon

Εε

Used in Greedy-Epsilon algo for multi-armed bandit problems

Error margin in floating point comparisons.

Zeta

Ζζ

Eta

Ηη

Theta

Θθ

Iota

Ιι

Kappa

Κκ

Lambda

Λλ

Mu

Μμ

Nu

Νν

Xi

Ξξ

Omicron

Οο

Pi

Ππ

Rho

Ρρ

Sigma

Σσς

Tau

Ττ

Upsilon

Υυ

Phi

Φφ

Chi

Χχ

Psi

Ψψ

Union

∪

Aleph

ℵ

Symbol for cardinal numbers. ℵ is pronounced as Aleph-null.

Empty Set

∅

Such That

Commonly represented as a colon, :

Example, D={x^2|x ∈ N, x >=1, x <= 4}. This reads D is the set of all x^2 SUCH THAT: 1) x is a natural number; 2) x is greater or equal to 1; 3) x is less than or equal to 4.

Intersection

∩

Subset

⊂ or ⊆

e.g., if A = {1,4,9} and B = {1,4}, then B ⊂ A (B is a subset of A).

Belongs To

∈

∉ Means “not belong”.

To say that 1 belongs to S, we write 1 ∈ S.

e.g., if A = {1,4,9} and e = 4, then we say e∈A, meaning “e belongs to A”. However, one would not say e⊂A – e is a single element, not a set. Similarly, if B = {1,4}, one would not say B∈A or “B belongs to A”, as B is a set not a single element.

Complements

Difference between two sets

Relative Complement

A\B means objects that belong to A and not to B. i.e., {1,2,3}{3} == {1,2}

Omega

Ωω

P(A|B)

The likelihood of event A occurring given that B is true.

P(A^C)

The probability that A doesn’t happen

Precision Recall

Precision = probability that some retreived doc is relevant; Recall = probability that some relevant doc was retreived.

Narrow Integration Tests

exercise only that portion of the code in a service that talks to a separate service; uses test doubles of those services, either in process or remote; thus consist of many narrowly scoped tests, often no larger in scope than a unit test (and usually run with the same test framework that’s used for unit tests)

Broad Integration Tests

require live versions of all services, requiring substantial test environment and networ access; exercise code paths through all services, not just code responsible for interactions

Balanced Binary Search Tree

For example, red-black tree or AVL tree.

Natural Numbers

ℕ

double-struck N

Cardinal numbers,

Complex Numbers

ℂ

double-struck C

Related

https://en.wikipedia.org/wiki/Blackboard_bold