iiinova — where your digital self keeps evolving

Santiago @svpino.com · 30/09/2025

Gemini's documentation says that "responses for a given prompt are mostly deterministic, but a small amount of variation is still possible." I'm not sure about you, but "mostly deterministic" is not the same as "deterministic." 9/9

030

Santiago @svpino.com · 30/09/2025

So, in practice, the LLM you are calling is non-deterministic, even at temperature = 0. OpenAI's documentation tells us that we can expect "(mostly) deterministic outputs across API calls." 8/9

110

Santiago @svpino.com · 06/01/2025

This guy freaked out when he tried a model much dumber than GPT 3.5. He ended up losing his job. He was an engineer and thought the model was sentient. Imagine what will happen when 99.99% of the world population uses one of these models for the first time.

4120

Santiago @svpino.com · 29/12/2024

This is a 10x move here. One of my favorite ways to leverage models.

2110

Santiago @svpino.com · 25/12/2024

Midwits with a chatbot will never build better software than you. These are two of my all-time favorite Software Engineering books: 1. The Pragmatic Programmer 2. Clean Code Read them in 2025. They will pay off for years to come.

3461

Santiago @svpino.com · 22/12/2024

This MIT class is still the best way to learn Linear Algebra. It's free. Gilbert Strang is one of those generational professors. The type of person that will leave a positive mark on you. ocw.mit.edu/courses/18-...

4825

Santiago @svpino.com · 20/12/2024

“This is not merely incremental improvement, but a genuine breakthrough” François Chollet about the just released OpenAI o3 model.

3222

Santiago @svpino.com · 19/12/2024

You can opt out of Visual Studio Code's Copilot from using your data to train their models. Works for free and pro users. This option is enabled by default. (You'll find this in your Copilot settings.)

3329

Santiago @svpino.com · 18/12/2024

Boomer developers are not going to make it.

230

Santiago @svpino.com · 16/12/2024

You can integrate uv with your shell to enable autocompletion on the terminal! The good stuff keeps getting better! Take 5 minutes and look into uv. I'm willing to bet you'll like it a lot.

2304

Santiago @svpino.com · 13/12/2024

Game-changing upgrade: Start using an automatic formatting tool that's opinionated. I haven't found one tool for Python that's better than black. It's fast, and I love the way it formats my code.

3221

Santiago @svpino.com · 11/12/2024

Most companies are still experimenting with AI. We are still far, far away from finding product-market-fit for all of the recent AI progress we've made. Talk to people outside your online bubble: 99% of AI applications still use boosted trees and convolutional neural networks.

260

Santiago @svpino.com · 10/12/2024

List comprehensions are one of the most elegant features of Python. I understand they look weird for people new to the language, but they are Pythonic and everyone uses them. Attached you'll see two examples. The second one is the way to go.

Two sample Python snippets computing the first 10 squares starting at 0. The first example uses a for loop. The second examples using a list comprehension.

4222

Santiago @svpino.com · 08/12/2024

I used to think that writing comments on your code showed a lack of ability. Your code must be self-explanatory. Comments are a crutch in many ways. They are hard to maintain, go out of sync easily, and often offer redundant information. 1/3

6161

Santiago @svpino.com · 05/12/2024

This is how Large Language Models work.

0100

Santiago @svpino.com · 03/12/2024

Using a calculator to review my son’s math is so slow. I can now ask ChatGPT to review the answers. It took me 10 minutes before. It now takes me 10 seconds.

6240

Santiago @svpino.com · 02/12/2024

We need a new word to represent "next level of hype."

4121

Santiago @svpino.com · 01/12/2024

What’s happening? Who is David Mayer and why ChatGPT blows up when you ask?

3120

Santiago @svpino.com · 30/11/2024

Best looking keyboard in 2024 amzn.to/3VfRFeb

7772

Santiago @svpino.com · 29/11/2024

This is probably one of the little Python tricks I use the most: `dir` returns everything that's available from an object/library. Super convenient when you don't know what exactly you can do with the result of a function, or don't remember the name of a specific property.

8621

Santiago @svpino.com · 28/11/2024

A 10x improvement to any customer-facing application: Start using semantic search in addition to your regular full-text searches. This is a PostgreSQL query using OpenAI's embeddings to find relevant articles. Results are magic.

2372

Santiago @svpino.com · 27/11/2024

Chrome doesn't like your site:

210

Santiago @svpino.com · 27/11/2024

If we change the activation function from "Linear" to "ReLU", the network will find a solution! Same network structure. Different activation function. ReLU introduces the non-linearity we need for the network to become much more powerful. 6/7

140

Santiago @svpino.com · 27/11/2024

The same network, however, can't solve the second problem. Look at the output here: it's all messed up. We can't solve this problem with a single line, so we need to configure this network to find non-linear solutions. 5/7

110

Santiago @svpino.com · 27/11/2024

A network with 2 hidden layers of 4 neurons finds the correct solution for the first problem in a few iterations. I used a "linear" activation function. Look at the output: The network created a single dividing line to separate the classes—that's all it can do with a linear activation. 4/7

210

Santiago @svpino.com · 27/11/2024

Here is a more complex problem. You have the same two orange and blue classes, but now, you can't use a simple line to separate these classes. Let's try to solve these two problems using a neural network: 3/7

110

Santiago @svpino.com · 27/11/2024

Here is a simple problem. You want to classify the orange and the blue data. Fortunately, we can separate both classes using a line, so this shouldn't be a big deal. This is a "linearly separable" problem. 2/7

110

Santiago @svpino.com · 26/11/2024

The best joke of 2024.

6453

Santiago @svpino.com · 24/11/2024

It's heartbreaking how many smart people have been led to believe leetcode is the only way to interview and find good candidates. I said last week, and I'll say it again today: If you base your interviews on leetcode questions, your interview process is a mediocre faang-wanabe.

3331

Santiago @svpino.com · 24/11/2024

Of all the different ways I use ChatGPT, one of my favorite use cases is asking it to analyze some text and then ask me questions about it. This is ideal for studying a topic. I'd have killed to have this back in college.

6351

Santiago @svpino.com · 24/11/2024

When was the last time you felt like this about a class you took? This is the feedback from my last cohort. I teach applied Machine Learning, and I do whatever it takes to blow my students away. The next cohort is in February. Today's price is the lowest it will ever be. ml.school

0151

Santiago @svpino.com · 21/11/2024

The confusion matrix is probably one of the most useful tools in machine learning. It's so simple to put together, yet so powerful. At a glance, you can determine the number of TP, FP, TN, and FN of a classification model, and then compute the model's accuracy, precision, recall, and f1-score.

0340

Santiago @svpino.com · 21/11/2024

Yeah, AI will definitely automate away a few Software Engineers.

3140

Santiago @svpino.com · 19/11/2024

I have never seen anybody ratio Musk. The replies to this post are horrendous. I read about 200 of them, and could not find a single person that agreed on speeding up legal, qualified immigration. Incredible. Here is the link to the original post: x.com/elonmusk/sta...

5111

Santiago @svpino.com · 16/11/2024

If you want to teach yourself machine learning, here are some of my favorite books.

4954

Santiago @svpino.com · 16/11/2024

X systematically boosted Elon Musk and right-leaning accounts (in detriment of left-leaning accounts) starting in July 2024, after Musk endorsed Trump. This paper is 👀

4285

Santiago @svpino.com · 15/11/2024

I just ate a chocolate from the hotel minibar and I’m now afraid they will bankrupt me.

1120

Santiago @svpino.com · 15/11/2024

We now have the first "AI Machine Learning Engineer." It's a multi-agent system that uses multiple models to aid with the entire ML lifecycle. Attached, you'll see the results on MLE-bench, a benchmark for measuring how well AI agents perform in ML engineering tasks. Neo comes on top.

1131

Santiago @svpino.com · 15/11/2024

My health insurance for 2025: $23,075 That covers my wife, two kids, and myself. The plan is not bad, but it's not great: • $5,000 deductible • $7,000 max out of pocket We tried to shop for cheaper options and they all had many restrictions in terms of doctors and coverage.

570

Santiago @svpino.com · 14/11/2024

No, Large Language Models didn't achieve the Kaggle Grandmaster level. It's unfortunate researchers have to resort to clickbait and misleading titles to get people to read their papers. Read the paper, and you'll see what's going on:

160

Santiago @svpino.com · 13/11/2024

Low-rank adaptation (LoRA) is truly awesome: 1. Replace a huge matrix with two (much) smaller matrices 2. Fine-tune the smaller matrics instead of the large one 3. Plug the fine-tuned adapter into a large model on the fly 4. Way faster and more efficient This is pure magic.

060

Santiago @svpino.com · 06/11/2024

A free, online, hard-core Machine Learning book! If you are interested in understanding how Machine Learning algorithms work, this is for you. Great resource if you are one of those who cares about how the magic happens. dafriedman97.github.io/mlbook/conte...

1192

Santiago @svpino.com · 04/11/2024

Rule #1 of Machine Learning: Do not start with Machine Learning.

021

Santiago @svpino.com · 29/10/2024

Python is now #1, like it should be.

050

Santiago @svpino.com · 17/05/2023

I can't dismiss this idea: I've read several anecdotes from researchers who can't explain some emergent behavior of large language models. These models solve problems they shouldn't be able to solve, and nobody can explain how they do it.

140

Santiago @svpino.com · 15/05/2023

Bard comes with something that ChatGPT doesn't have: Internet access and multilingual support. OpenAI answered by giving every pro user access to plugins. Everyone can now connect ChatGPT to the Internet. Two powerful AI tools head to head.

110