AI ML

HOY
- Tipos de modelos
- Como se entrenan
- GPT3, 3.5, 4
- Prompt engineering
- Danger

- Porque
- Cosas chulas
DISCLAIMER
- Muchas cosas estan sobre simplificadas
- Es una charla introductoria
- No soy un experto en IA
- Cambia constantemente
Porque es importante?

Porque es importante?
Aceleracion

Porque es importante?
La revolucion industrial
Manufacturar -> Fabrica
La revolucion IA
Oficinas -> ????

Ya existe

Ejemplo

Generador de prompt de midjourney
Midjourney a imagen
Profit
Los ultimos 10 dias
Tipos de modelos

Imagen
Stable diffussion
MidJourney
Dall-E
Craiyon
...

Codigo
Chat GPT
Stability.ai
Tabrine
CoPilot
...

Video
ModelScope
Synthesia
deepbrain ai
Texto
LLama
GPT-Neox
GPT-4

Donde estan

Los modelos de ML
IA > ML

INPUT
OUTPUT
Los modelos de ML
Dentro de la caja
Los modelos de ML
Dentro de la caja

Los modelos de ML
Dentro de la caja

Como entrenar a tu dragon

Escoges tu modelo

Le das datos de comer

Ves el futuro

Fine tunning
ANY MODEL
NEW DATA
YOUR MODEL

O pasas de todo y usas GPT4 con contexto

GPT4
"Generative Pre-trained Transformer" (Transformador Generativo Preentrenado).
Generativo: crea nuevos datos basado en patrones aprendidos.
Preentrenado: se entrena en un gran conjunto de datos antes de ajustarse a tareas específicas.
Transformador: arquitectura subyacente para procesar y comprender secuencias largas de datos.
GPT4
Es un LLM (Large Language Model)
SOLO predice la siguiente palabra
pero tiene algo llamado atencion o contexto
Piensa?

Porque es la hostia?
Wed, 6 Dec 2017 03:30:32

Embeddings

Embeddings
Esto esta basado SOLAMENTE en la prediccion de la siguiente palabra con los datos de entrada que se ha ENTRENADO
Si usamos diferentes datasets para entrenar desde CERO chatgpt
Libros del siglo XIX -> embeddings de libertad
Libros del siglo XX -> embeddings de libertad
Da el mismo valor?
Embeddings

Parametros
GPT3 -> 175 billion parameters

GPT4 -> 1 trillion parameters
Prompt engineering

Prompt engineering
Resume este texto de debajo en una frase y con una lista de los 5 puntos mas importantes.
Text: """
At GitHub, our mission has always been to innovate ahead of the curve and give developers everything they need to be happier and more productive in a world powered by software. When we began experimenting with large language models several years ago, it quickly became clear that generative AI represents the future of software development. We partnered with OpenAI to create GitHub Copilot, the world’s first at-scale generative AI development tool made with OpenAI’s Codex model, a descendent of GPT-3.
GitHub Copilot started a new age of software development as an AI pair programmer that keeps developers in the flow by auto-completing comments and code. And less than two years since its launch, GitHub Copilot is already writing 46% of code and helps developers code up to 55% faster.
But AI-powered auto-completion is just the starting point. Our R&D team at GitHub Next has been working to move past the editor and evolve GitHub Copilot into a readily accessible AI assistant throughout the entire development lifecycle. This is GitHub Copilot X—our vision for the future of AI-powered software development. We are not only adopting OpenAI’s new GPT-4 model, but are introducing chat and voice for Copilot, and bringing Copilot to pull requests, the command line, and docs to answer questions on your projects.
With AI available at every step, we can fundamentally redefine developer productivity. We are reducing boilerplate and manual tasks and making complex work easier across the developer lifecycle. By doing so, we’re enabling every developer to focus all their creativity on the big picture: building the innovation of tomorrow and accelerating human progress, today.
Let’s jump in.
Want to see what’s new? Discover GitHub Copilot X—our vision for the future of AI-powered software development. Learn more >
A graphic showing how many developers and companies have already used GitHub Copilot and how it's helping improve productivity and happiness.
Many developers and companies have already used GitHub Copilot, and it’s helping improve productivity and happiness.
A new AI-powered developer experience 🧑💻
A ChatGPT-like experience in your editor with GitHub Copilot Chat: We are bringing a chat interface to the editor that’s focused on developer scenarios and natively integrates with VS Code and Visual Studio. This does far more than suggest code. GitHub Copilot Chat is not just a chat window. It recognizes what code a developer has typed, what error messages are shown, and it’s deeply embedded into the IDE. A developer can get in-depth analysis and explanations of what code blocks are intended to do, generate unit tests, and even get proposed fixes to bugs.
GitHub Copilot Chat builds upon the work that OpenAI and Microsoft have done with ChatGPT and the new Bing. It will also join our voice-to-code AI technology extension we previously demoed, which we’re now calling GitHub Copilot Voice, where developers can verbally give natural language prompts.
Sign up for the technical preview >
Copilot for Pull Requests: You can now sign up for a technical preview of the first AI-generated descriptions for pull requests on GitHub. This new functionality is powered by OpenAI’s new GPT-4 model and adds support for AI-powered tags in pull request descriptions through a GitHub app that organization admins and individual repository owners can install. These tags are automatically filled out by GitHub Copilot based on the changed code. Developers can then review or modify the suggested description.
Enroll your repository in the technical preview >
This is just the first step we’re taking to rethink how pull requests work on GitHub. We’re testing new capabilities internally where GitHub Copilot will automatically suggest sentences and paragraphs as developers create pull requests by dynamically pulling in information about code changes.
We are also preparing a new feature where GitHub Copilot will automatically warn developers if they’re missing sufficient testing for a pull request and then suggest potential tests that can be edited, accepted, or rejected based on a project’s needs.
This complements our efforts with GitHub Copilot Chat where developers can ask GitHub Copilot to generate tests right from their editor—so, in the event a developer may not have sufficient test coverage, GitHub Copilot will alert them once they submit a pull request. It will also help project owners to set policies around testing, while supporting developers to meet these policies.
Get AI-generated answers about documentation: We are launching GitHub Copilot for Docs, an experimental tool that uses a chat interface to provide users with AI-generated responses to questions about documentation—including questions developers have about the languages, frameworks, and technologies they’re using. We’re starting with documentation for React, Azure Docs, and MDN, so we can learn and iterate quickly with the developers and users of these projects.
"""GitHub avanza en IA para desarrollo de software con Copilot X:
1️⃣ Copilot escribe el 46% del código y aumenta velocidad en un 55%.
2️⃣ Introducción de GPT-4, chat y voz para Copilot.
3️⃣ Copilot en pull requests y línea de comandos.
4️⃣ Generación de pruebas y advertencias automáticas.
5️⃣ Respuestas de IA para preguntas en documentación.Resumen de este tocho
Output
Prompt engineering
I am a highly intelligent question answering bot.
If you ask me a question that is rooted in truth, I will give you the answer.
If you ask me a question that is nonsense, trickery, or has no clear answer,
I will respond with "Unknown".
Q: What is human life expectancy in the United States?
A: Human life expectancy in the United States is 78 years.
Q: Who was president of the United States in 1955?
A: Dwight D. Eisenhower was president of the United States in 1955.
Q: Which party did he belong to?
A: He belonged to the Republican Party.
Q: What is the square root of banana?
A: Unknown
Q: How does a telescope work?
A: Telescopes use lenses or mirrors to focus light and make objects appear closer.
Q: Where were the 1992 Olympics held?
A: The 1992 Olympics were held in Barcelona, Spain.
Q: How many squigs are in a bonk?
A: Unknown
Q:
Q: Who won more medals in the 1996 olympic games?
A: The United States won the most medals in the 1996 Olympic Games with 101 medals.Q&A
Output
Prompt engineering
A table summarizing the fruits from Goocrux:
There are many fruits that were found on the recently discovered planet Goocrux.
There are neoskizzles that grow there, which are purple and taste like candy.
There are also loheckles, which are a grayish blue fruit and are very tart,
a little bit like a lemon.
Pounits are a bright green color and are more savory than sweet.
There are also plenty of loopnovas which are a neon pink flavor and taste like cotton candy.
Finally, there are fruits called glowls, which have a very sour and bitter
taste which is acidic and caustic,
and a pale orange tinge to them.
| Fruit | Color | Flavor |
Q: Who won more medals in the 1996 olympic games?
A: The United States won the most medals in the 1996 Olympic Games with 101 medals.Clasifica datos no estructurados
Output
Prompt engineering
Given this conversation:
"""
- Irene: 👏🏻👏🏻👏🏻
- Pipo: Han sido un poco raras
- Alejandro: Claramente otros barrios mejor
- Alejandro: Para ser pozuelo es bastante cutre
"""
Output a table with participants their role (how they behave in the conversation)
and who they reply to most (one person)
In this format
| Name | Role | Interacts with most || Alejandro | Jokester | Miguel Angel Brualla de Cruz |
| Carlos Rodríguez Centeno | Questioner/Inquirer | Miguel Angel Brualla de Cruz |
| Dario | Congratulator | Maca Pablo |
| Gabi Mariquito | Jokester | Pipo |Quien hace que en conversacion
Output
Prompt engineering
I want you to act as a linux terminal.
I will type commands and you will reply with what the terminal should show.
I want you to only reply with the terminal output inside one unique code block,
and nothing else. do not write explanations.
do not type commands unless I instruct you to do so.
When I need to tell you something in English,
I will do so by putting text inside curly brackets {like this}. My first command is pwd/home/user
ls
Desktop Documents Downloads Music Pictures Public Templates Videos
Act as a terminal
Output
Prompt engineering
I want you to translate the sentences I wrote into emojis.
I will write the sentence, and you will express it with emojis.
I just want you to express it with emojis.
I don't want you to reply with anything but emoji.
When I need to tell you something in English,
I will do it by wrapping it in curly brackets like {like this}.
My first sentence is "Hello, what is your profession?"👋🏼🤔💼Emojii conversation
Output
Prompt engineering
Meta question
(que pregunto para parecer list@)
-> Explain to me the theory of ideas of Plato
The theory of ideas, also known as the theory of Forms, is one of the most famous and influential ideas developed by the ancient Greek philosopher, Plato.
According to Plato, the physical world that we can see and touch is not the ultimate reality, but rather a flawed copy or imitation of a higher reality that exists beyond the world of appearances. This higher reality is made up of eternal and unchanging "Forms" or "Ideas" that serve as the true objects of knowledge.
-> what would be a good interesting question about this?
How does Plato's theory of ideas or Forms challenge our understanding of the relationship between knowledge and perception, and what implications does this have for our understanding of reality?Prompt engineering
Todo es un lenguaje
I want you to act as a Graphviz DOT generator, an expert to create meaningful diagrams.
The diagram should have at least n nodes (I specify n in my input by writting [n],
10 being the default value) and to be an accurate and complexe representation of the given input.
Each node is indexed by a number to reduce the size of the output, should not include any styling,
and with layout=neato, overlap=false, node [shape=rectangle] as parameters.
The code should be valid, bugless and returned on a single line, without any explanation.
Provide a clear and organized diagram, the relationships between the nodes have to make sense
for an expert of that input.
My first diagram is: "The water cycle [8]".I want you to generate a music composition, a house/techno song with chiptune sounds,
do it by creating a snippet of code that can run on sonic pi.
The song should have several tracks that complement each other like bass, synth, beat, etc.
It should last 1 minute.Musica
DANGER

Como es GPT
como un bonaerense o tu cuñado
en la cena de navidad

Hallucination

Bias

Context limit

4k -> 8k -> 32k
Near future
[Alice] Quiero este trabajo (x)
tu IA prepara el CV y aplica
su IA lee el CV y resume
de estos 3 [Alice, Bob, Mike] escoge 1
IA publica posicion de trabajo (x)
Cosas chulas

Cosas chulas

Cosas chulas

Cosas chulas

Cosas chulas

Cosas chulas

Cosas chulas

Questions?

AI/ML
By Alejandro Vidal Rodriguez
AI/ML
Micrometer + prometheus + grafana + alertmanager
- 58