AI Daily Grind // AI news, freshly ground daily

Learn · Models and how they work

What is Weights?

Illustration: Weights
Simply put, like I'm 5 years oldWeights are like the knobs on a giant music mixer. During training, the AI turns millions of little knobs until the music sounds right, and those knob positions are what it "knows."

The slightly longer version

An AI model is mostly a huge list of numbers called weights. Each weight is a tiny dial that says how much one piece of information should count. During training, a computer shows the model lots of examples and slowly adjusts the weights, one tiny step at a time, until its answers get better. When you type a question, your words become numbers that pass through all these weights, and the answer comes out the other side.

A small model like Qwen3.5 4B has about 4 billion weights. The "4B" in the name means 4 billion. When a company publishes the weights, we call it an open-weight model. Anyone can download it, run it on their own computer, study it, or change it. When weights stay secret, you can only use the model through the company's app or service.

How weights turn a question into an answer

  1. Start with random numbersA new model begins with billions of weights set more or less by chance.
  2. Training adjusts themThe model sees many examples and its weights are nudged a little after each mistake.
  3. Weights are savedThe final numbers are stored in a file, and that file is the model.
  4. Your question goes inYour words become numbers that flow through all the weights.
  5. An answer comes outThe math through the weights produces the next words of the reply.
Why it matters

This week, someone built a small working computer processor directly inside the weights of Qwen3.5 4B, a project called Qwenputer. It shows that weights are not magic. They are just numbers, and clever people can shape them to do very precise jobs. For you, it is also a question of control. Whoever holds the weights decides who can use a model, check it, or change it.

What to watch out for

Qwenputer is a clever experiment, not a product. The author thinks it might help with tool use, but that is not proven. More generally, even experts cannot read weights and fully explain why a model gives a certain answer. Open weights also do not mean the model is checked or safe: this week there was a dispute about a Qwen model phoning home, so test downloaded models before you trust them.

Where it showed up

This notion came up in Smarter hackers, sneakier agents, and a public that is losing patience. More lessons: Learn.

Sources

Written on 2026-10-29. This explanation was accurate on that day. AI changes fast, so check the sources and the date before you rely on it.

Get The AI Daily Grind by email, every morning, in simple words.