One of the dervishes said to a Bedouin: Shall I teach you three virtues for which people will love you? The Bedouin answered: No...

Instead of LLM training eras think of it like this: Pretrain: learn the language Mid-train: get closer to the goal (domain/long context) SFT: teach instructions Pref/RL: steer toward the right answers, cut down on nonsense LoRA/QLoRA: not a new era—an inexpensive speed-up for

Today’s topic: File I/O: with + encoding + reading line by line Form this sentence in your mind: “If a resource is opened, it must be guaranteed to be closed.” Quick explanation: with opens a file safely and automatically closes it when you’re done. If there’s a syntax pattern:

My goal was to get past the 1k barrier by the end of 2025 and thank God I did :)

Stop stepping the scheduler however you feel like it. Step it after optimizer.step() (especially OneCycleLR, CosineWarmRestarts). The wrong order shifts the LR curve by 1 update; then you spend 3 days tuning while chasing a bug. Stop slapping weight_decay on everything. Use

Today’s topic: lambda in Python Form this sentence in your head: “I need a tiny function… right here, right now.” That’s why the syntax is: lambda inputs: expression So first you say what comes in (parameters), then what comes out after : (the expression). No def, no name, no

Today's topic: map in Python Form this sentence in your head : I have a function, and I have a list. Apply this function to EVERY element of the list. That's why the syntax is "map(what_I_will_do, what_to_do_it_on)" So first you say what you are going to do (the function),

Hey everyone, if you’re new to machine learning as of today, I created a repo for you and I’ll constantly be adding the things you need to know at the start, real projects included; as I do this, I’m also developing myself, let me repeat that. I tried to keep the comments in the

Love this!

Aurum nostrum non est aurum vulgi - Alchemy and Alchemists

What Is Multiple Regression? You can think of multiple regression as the “upgraded” version of linear regression that lets you make predictions not by looking at just one cause, but by taking a bunch of causes into account at the same time; for example, when predicting a house

As a mechanical engineer, I have to say Gemini 3.0 Pro is absolute gold for spotting errors in my drawings. I put GPT 5.1 to the test as well, and it caught every single mistake Gemini did accuracy-wise they are neck and neck. But the real gamechanger here is speed. I can’t speak

One does not become enlightened by imagining figures of light, but by making the darkness conscious - C.G. Jung, Alchemical Studies

Your titles hold sway only until the shroud. Death is the most patient artist; he always applies the final stroke. - “Danse Macabre” woodcut by Hans Holbein, 1526.

That sky, which all perceive as the final frontier, is in truth merely a thin veil that only the brave can part. Truth reveals itself solely to those who dare to peer beyond that comfortable enclosure This famous illustration is from Camille Flammarion's 1888 work, L'Atmosphère:

Today’s topic is linear regression in machine learning ; If you’re using it like “Linear regression? Just call LinearRegression().fit(X, y) and that’s it…” but you don’t really understand what it’s doing, this thread is exactly for you. Here, we’ll explain the formula in

There’s a popular myth going around on the internet, you’ve probably seen it as well: some people say “LLMs are a total black box, nobody knows what’s going on inside them,” and some even push it all the way to talk of “hidden intelligence, consciousness, magic.” Told like this,

Hey everyone, - I’ve just released my new Chrome extension. You can download photos and videos, export the list of people who viewed your story, and also export your DMs. As far as I know there’s no free extension that gives you a story viewers list right now, so I hope it’s

"Mathematics is the alphabet with which God has written the universe." – Galileo Galilei In ancient teachings and occultism, the spiral is not merely a shape, but the very breath of the universe. They stand as geometric proof of the Hermetic principle: "As above, so below." The

If you’re dealing with object detection right now, you’re probably getting lost in the “YOLO, Faster R-CNN or DETR?” debate. In reality, it’s not about individual models; once you clarify which paradigm you’re playing in, things get much easier. The usual reflex is this: you

I've been writing down every feature I’ve been missing for a long time and I’m putting them all into an Instagram extention. Born from my own needs, a plugin that will be useful to everyone is on the way.