Saved

Magical courses made just for you.

The best article in this year so far. The instruction tune process makes LLMs sound smart, but they're not really that clever.

It's like what's happening in research - papers filled with complex notations and abstract concepts just to seem impressive.

Instead of just solving problem straightforward, like in the article point out that when using LLMs to find the vulnerabilities, if the code looks good, just tell it's fine. But LLMs always try to highlight any potential issues, when everything's just works.

A small tool can improve overall agentic capabilities.

The end-to-end platform for building world-class AI apps.

Magic is working on long context and inference-time compute, they are focusing on coding purposes and building models from the ground up.

Finding the simplest solution possible, only increasing complexity if needed.

Prompt processing is cheap, generating tokens is slow and expensive.

An engineer at Gitlab accidentally deleted their production database, and they couldn't recover it because their backup methods failed. They shared the root cause, how they handled the situation, and the lessons learned in a very transparent way.

The refactoring pull request of the Bluesky app, specifically how they plan to store data. They use a single SQLite database per user that stores their repository and private account state.

Beautiful artworks! I'm currently using them as my wallpaper on both my Mac and iPhone.

"...many companies will never grow large enough or achieve enough scale for this complexity to be beneficial".

Right tools for the right job, isn't it?

"Then when the kids all left the nest, I looked in the mirror and asked myself, who am I, really?"

Is the planning we do all the time worth it, or is it just an illusion of control?