LLMs cannot reason.

Despite their impressive capabilities, all LLMs, including OpenAI o1, are still fundamentally limited by design constraints that make them incapable of true, open-ended reasoning.

Let's break it down. 🧵 (1/5) Reason 1: Stochastic Sampling.

LLMs rely on probabilities to pick the next token. Even when you fix the temperature, randomness is still built into the language modeling paradigm.

But logic is anything but random.

What is Machine Learning?

Here's a 3 min intuitive introduction explaining why this is the most powerful paradigm shift for conventional software development.

No math, no code, just intuitions. Let's dive in. Conventional software is about automating stuff.

A client comes up with a problem, and to solve it, as a software developer, you must understand very precisely how a human would do it.

Then you can proceed to automate that process.

Clustering is a process for discovering relationships between objects, by placing them in different groups according to how similar they are with each other.

❓Given a set of objects, is there a natural, unbiased way to cluster them?

Meet the ugly duckling theorem.

🧵 1 of 20 Say we have three objects, two swans 🦢🦢 and an ugly duckling 🦆.

Obviously, the natural way to cluster them is by placing the two swans together and the duckling in a different group, right?

Well, it depends on which features you choose to look into.

2 of 20

I've spent the last couple of years disrupting traditional software companies with machine learning and data science ideas directly out of my group's core research.

I've found that most issues arise from three critical areas.

Here's what I've learned...

🧵 1/24 Most of the obstacles I've seen can be grouped in one of the following three categories:

1️⃣ the language
2️⃣ the development process
3️⃣ the expected results

Let's tackle them one by one.

2/24

AutoML is a growing subfield of machine learning, that aims to automate some of the most boring and time-consuming parts of designing, training, and deploying a machine learning pipeline.

Here are 10 open source AutoML tools you can start using today:👇 ⚙️auto-sklearn

Probably the most popular AutoML system, it sits on top of everyone's favourite ML framework, scikit-learn, and gives you a black-box AutoML wrapper that abstracts away most of scikit-learn's estimators.

🔗github.com/automl/auto-sk…

Have you heard about P vs NP?

It's probably the most important theoretical question in computer science, and it sounds weirdly abstract.
But deep down, it has a very intuitive explanation.

If you have heard of this and want to learn a bit more, read on...

🧵👇 Computer Science is all about finding clever ways to solve difficult problems.

We have found clever algorithms for a bunch of them: sorting stuff, finding shortest paths, solving equations, simulating physics...

But some problems seem to be way too hard 👇

If you're looking for an easy way to introduce more machine into your products and services, AutoML is a good bet.

✨ Here's a short (and incomplete) list of open source and commercial AutoML systems you can start using today:

autogoal.github.io/survey/systems 1- AutoSklearn (automl.github.io/auto-sklearn/m…)

A drop-in replacement for scikit-learn that can train classifiers or regressors automatically. Based on Bayesian Optimization, and production-ready.

Hi 👋 I'm Alejandro from Havana, Cuba 🇨🇺

By day, I'm a CS college professor and AI researcher. By night, I come to Twitter to explain things as simple as possible, but not simplistically.

✨ Follow me if you're interested in technical and philosophical aspects of AI and ML. 🧵 I like to write threads, once or twice a week, on topics related with AI and CS in general.

✍️ I also like to go deeper on some topics with short essays and technical guides. You can check them all at apiad.net.

Here's my Twitter schedule: 👇

Hey folks 🖖!

😎 Happy Sunday to everyone! How about we make a round of presentations? Let's meet and follow as many cool people as possible today!

Tell me who you are and why should we follow you. I'll start 👇 👋 I'm Alejandro from Havana, Cuba 🇨🇺

By day, I'm a CS college professor and AI researcher. On the nights, I enjoy explaining things as simple as possible, but not simplistically.

✨ Follow me if you're interested in AI, machine learning, and computer science in general.

Solving a machine learning problem requires more than choosing the right model.

One of he most challenging and possibly critical design decisions in any ML process is figuring out which evaluation metrics to use.

Let's talk about choosing the right metric in ML.

🧵👇 ✏️ This is the last of 5️⃣ threads I shared with you this week on foundational topics in machine learning.

If you want a recap, or haven't seen the rest, check out the pinned tweet @AlejandroPiad.

Hey folks 🖖!

This is the second thread in a series on the foundations of Machine Learning.

❓ Today, I want to start discussing the different types of ML flavours we can find.

This is a very high-level overview of the different ML paradigms.

🧵👇 1/18 Last time we talked about how Machine Learning works.

👉 Basically, it's about having some source of experience E for solving a given task T, that allows us to find a program P which is (hopefully) optimal w.r.t. some metric M.

Hey folks 🖖!

This is first thread in a Twitter series about the foundations of Machine Learning. To begin, I want to answer this simple question.

❓ What is Machine Learning?

This is my preferred way of explaining it.

🧵👇 1/22 Machine Learning is a computational approach to problem-solving with four key ingredients:

1️⃣ A task to solve T
2️⃣ A performance metric M
3️⃣ A computer program P
4️⃣ A source of experience E

Hey folks 🖖!

😎 Happy Sunday to everyone! How about we make a round of presentations? Let's meet and follow as many cool people as possible today!

Tell me who you are and why should we follow you. I'll start 👇 I'm Alejandro 🇨🇺. I'm passionate about explaining things as simple as possible, but not simplistically.

🎓 I teach programming, compilers, and AI in the computer science major at the University of Havana.

For most of our history on Earth, human-to-human communication has been ephemeral.

We didn't need to care too much for what we said, as most of it wasn't supposed to be remembered.

This has changed in the last 30 years. We now have, for the first time ever, the technology to store every one of our thoughts forever.

Whatever we say in public can be recalled years later and used to support (or contradict) any claims about ourselves, our past, our character.

Hey folks, today is a Theory Thursday 🧐!

The perfect excuse to share with you random bits of theory from some dark corner of Computer Science and make it as beginner-friendly as possible.

Today I want to talk about *Algorithmic Complexity*.

🧵👇 1/22 To get started, take a look at the following code. How long do you think it will take to run it?

Let's make that question more precise. How long do you think it will take to run it in the worst-case scenario?

👇 2/22

Grad school is hard. You need to be both open to explore new and crazy ideas and able to laser-focus on one of them at a time.

These are three healthy research habits I wish I had known when I started my PhD.

🧵👇 1/19 Grad school is one of the most intense moments in a researcher's career.

You have to go through a cycle of ideation, experimentation, and writing, as fast as you can, multiple times during those few years, which risks burning you out.

👇 2/19

Hey folks, today is a Technical Tuesday 🤓!

Let's talk about practical technologies that you can use today to improve your existing workflows.

In this thread I will tell you about *AutoML*

🧵👇 1/16 AutoML stands for *Automated Machine Learning*.

It encompasses a bunch of technologies and paradigms to gradually automate the process of creating machine learning solutions.

💡 AutoML is about raising the abstraction level in ML and reducing the grunt work.

👇 2/16

Hey folks, today is a Mindblowing Monday 🤯!

A day to share with you amazing things from every corner of Computer Science.

Today I want to talk about Generative Adversarial Networks.

🧵👇 1/11 🍬 But let's begin with some eye candy.

Take a look at this mind-blowing 2-minute video and, if you like it, then read on, I'll tell you a couple of things about it...

<>

👇 2/11

Most of the decisions that we remember making are which-decisions.

Which-decisions are the ones we have to make explicitely, when given two or more options, and it is impossible not to choose.

Which recipe to cook, which brand of phone to buy, which career to study.

👇🧵 1/7 Their explicit nature forces us to think and weigh their implications consciously before choosing.

We can make the wrong choice, sure. But, in hindsight, more often than not we tell ourselves that we chose the best we could with the information we had.

👇 2/7

The literature review is one of the most important, but often the most dread boring part of any research. I know, I've been there too.

In this thread, I will show you some techniques to make this process easier, and maybe even a little fun.

👇🧵 1/30 ⚠️ Disclaimer: this advice is based on personal experience from me and my close colleagues. I work in Computer Science, specifically in machine learning.

Every field is different, but I believe this advice is useful for anyone at least in quantitative fields.

👇 2/30

Convincing an audience to care for your idea, project, product, or business requires being persuasive when writing and talking.

In this thread you'll learn about the VSN-C framework, a simple way to structure a talk or writen piece for maximum persuasion.

🧵👇 1/14 Patrick Winston, legendary professor of AI at MIT, was a marvelous lecturer and communicator.

Part of his magic recipe was the VSN-C framework, a way of structuring an exposition or talk for maximizing the chance of convincing your audience.

This is how it works. 👇 2/14