Frequently asked questions

A free place to learn AI by building it. The learning part lives on its own site, learn.nanofuze.com: learning paths (goal-shaped routes such as AI from zero, building agents, how large language models work, robotics and physical AI), three full courses with code that runs, quizzes and spaced review, Here on nanofuze.com you find a Lab of tools, a glossary, and one news page that keeps you up to date with new models, prices, research and rules.

It is free. News, the Lab, the glossary and the rest of nanofuze.com need no account. The learning site (learn.nanofuze.com) asks you to sign in with your Google account; it receives your name and email address and nothing else. Your progress, notes, daily goal and review queue are saved in your own browser, not on your account, and are never sent to us. Clearing your browser's site data starts you again from the beginning.

If you are new to AI, take the learning path called AI from zero: seventeen short chapters that explain what AI, machine learning and language models are and how to use them well. If you can already code and want to build with language models, take Build AI agents. The learning site has a line for each kind of reader.

A course teaches one field in its own order, from the first idea to the last. A learning path picks the chapters you need, from any course, for one goal (for example, how large language models work) and groups them in stages, so you always know what to read next.

Each chapter is short, usually three to ten minutes. Code examples run, and the real output is shown beneath them. Quizzes inside chapters and questions at the end make you recall answers, and every question you answer joins a review queue that brings it back just before you would forget it. Explorables let you drag a slider and watch an idea change, and Nano, the assistant on the learning site, knows which chapter you are on and can explain it, quiz you or set an exercise.

Yes. Every Python example marked as runnable was executed with the libraries named on the course page, and the output shown is what it printed. A few short snippets, such as one that downloads pretrained weights, were only checked for syntax and are not marked as runnable. Each course says what was run and what was not.

Every hour. An automated process reads public data, compares it with the last reading and saves what changed. The Arena leaderboard is published about once a day, so ranking moves appear about daily.

It reads public data automatically: the OpenRouter model list for new releases and prices, the Arena leaderboard for the top 30 (people vote on blind answers), and Hugging Face for trending open models. A lab that appears in that data and is not in our directory gets a profile too, marked as found automatically. Our own notes on running costs, licences and retirements are written by hand and layered on top. If a source is down, the page shows our last hand-checked data and says so.

Nobody, and no language model. Each item is a difference between two readings of public data (a price that changed, a model that entered the top ten, a version that shipped), and its sentence is a template filled with the numbers. How important it is, from 1 to 5, comes from written rules that are published on the Data sources page.

From public data: OpenRouter's model list, the Arena leaderboard, Hugging Face (trending models and daily papers), the release feeds of the software that runs AI, and the United States Federal Register. The Data sources page lists each one, how often it is read and how each kind of item is made.

Each number is shown as the source publishes it, with a link to that source. Leaderboard scores come with their margin of error. We do not repeat a company's own claims about itself as facts, and an item says what the data shows, never why.

No. We do not read, copy or rewrite articles from other websites. Each news item links to the public data it was detected from.

They are now one page, AI news. It has the feed of everything that changed (with filters), price changes, new models, the live top 30, trending open models and the catalog of every notable model. The old addresses lead there, and each item keeps its own permanent link.

Yes. The AI news page has an RSS feed with the newest changes: nanofuze.com/signals/rss.xml.

Use the Corrections page. Tell us which chapter or news item and what is wrong, ideally with a link that shows it. Verified errors are fixed against the source and logged on that page.

Use the Send a tip page. Tell us which public data we should be watching, or what you would like a course to cover, and why it matters.

Yes, on the learning site (learn.nanofuze.com). Tap Ask Nano (or press Alt and V) and type, or tap the microphone and say things like "open the learning paths", "explain attention" or "what changed today?". Voice input needs a browser with speech recognition, such as Chrome, Edge or Safari. Firefox does not offer it, but typing always works.

Yes, on the learning site, if you turn it on. Open the assistant and switch on Hands-free, then say "Hey Nano" followed by your request. It is off by default, shows an indicator while it listens, pauses when you switch tabs or while it is speaking, and turns itself off after 30 minutes of quiet.

On the learning site, your browser's own speech service turns your voice into text (in Chrome that service is run by Google); Nanofuze never receives your audio. In hands-free mode your browser hears audio continuously while the switch is on, and Nanofuze only receives what you say after "Hey Nano". Your question is sent to our server so the assistant can answer it. We do not store what you ask.

You are welcome to link to any chapter, path or news item. Short quotations with attribution and a link are fine. Please do not republish the site wholesale. See the Terms of use for details.