Join Polaris

Join Polaris to use the free chapters and keep your prompts. Accounts, the market and the job board open later. During the trial, the forum is open to invited members. Every chapter is already free to read, with no account and no card.

Accounts

The forum is open to invited members, and they can sign in now. Course accounts open later. Leave your email and we will tell you when.

Forum sign-in Newsletter

Beginner

The promises and a three-day start

You can read every Polaris chapter today with no account and no card. The chapters are in English and Swedish and start from zero: what AI is, how to write to a chat model, and when to doubt an answer. The form will ask for your name, email, and a password, and it will take about a minute. Accounts are run by Hrafninn Industries AB in Kungsbacka, Sweden.

The sign-up form will make these promises. The chapters and the feed stay free to read. You start with three small steps: read one item in the feed, ask one question, and copy one prompt into any AI chat. You follow a three-day starter or one evening course. If the station is bought, a gift toward it gives a vote on which models run there, but no share, no claim on the machine and no reserved time. If the station is bought, founding members get their name on the chassis. The price of every chapter stays at zero.

Day one: read the chapter "What is AI?" and write one sentence you could say to a colleague. Day two: paste the starter prompt into a chat model you already use and keep the answer. Day three: write the question you still have, plus what you already tried.

For one evening, use Elements of AI from MinnaLearn and the University of Helsinki, a free course on what AI is and what it can and cannot do. The confirmation mail will ask you to confirm your address, repeat the three acts, and tell you how to stop further mail. Your name and email are personal data, information that can identify you.

My first three acts

I am new to AI. Using plain words, do three things. First, explain what a chat model is in five sentences, and mark any sentence that is only a guess, and name the kind of source that would settle it. Second, write one question I could post on a public AI forum after reading a beginner's chapter titled "What is AI?". The question must say what I already understand and what I tried. Third, write a short prompt I can reuse to check an AI answer against a source I name. Then give me a three-day plan with one small task for today, one for tomorrow, and one for the day after. Do not tell me to buy anything.

Intermediate

Save a job and publish a test

After you can get a useful answer from a chat model, the account will keep the work and let other readers repeat it. It will store three kinds of items. A saved job is a prompt plus the answer you decided to keep. A watch is a topic you want in the feed, such as Swedish-language models. A test post is a note someone else can run again.

Write a test post in four lines. Name the model. Paste the prompt. Paste the result. Say what you wanted, and quote any sentence you checked against a real page. If you change the prompt, save a new job so the old result stays tied to the old wording.

A project states a question, names material you have the right to use, and says what done means. A hackathon is the same thing with a fixed deadline, and you still work on your own laptop. If the station is bought, a larger run would go to a shared queue with the model name, the task, and a time limit written down first.

Practical Deep Learning for Coders, from fast.ai, is free if you already write some code, and it includes data ethics. The Hugging Face LLM course is free, has no ads, and expects Python. Early chapters show how to take a model, fine-tune it (train it further on your own examples), and share the result. That course text is under the Apache 2.0 license: reuse it if you give credit, link the license, and mark changes. Before a post goes in the feed, remove other people's personal data, passwords, and unpublished customer files.

Turn a chat into a test post

Help me write a public test note about an AI chat I already ran. Ask me for the model name, the exact prompt, the result, and what I wanted. Then format four lines labeled Model, Prompt, Result, and Check. In Check, quote any sentence I compared with a real page, and name that page. List any detail that is someone else's personal data, a password, or an unpublished customer file, and tell me to delete it before posting. Suggest one follow-up test that changes only one sentence of the prompt, so the two results can be compared. Keep the note short enough to paste into a forum.

Expert

Measurement and a vote

If the station is bought, expert membership covers work on it: a published measurement, and a vote on which models are installed.

Read the license before you propose a model. The Open Source Definition, from the Open Source Initiative, says open source software must meet a published list of criteria. The first is free redistribution: people may pass the software on without paying a royalty. Code and weights (the trained numbers inside a model) often use different licenses, so name both.

A queue job has three fields: model name, task, and a time limit you set beforehand. Inference means the model produces an answer. Fine-tuning means further training on examples you have the right to use. If the station is bought, founding members, whose gifts go toward it, get a vote on the next model and their name on the chassis, but no reserved time. If the station is bought, every account can join the general queue. Each proposal lists the license, whether Swedish text is in scope, the disk space required, and the measurement you will publish.

A measurement records the settings, the test set, the score, and the date. AI Sweden lets partner organizations test hardware and datasets in its technology infrastructure, and some of that partner work may cost. Gifts here go toward the station, and the written guides stay free to read. Mozilla's manifesto treats transparent community processes as support for participation and trust, which is why the vote and the measurement are public, and it treats privacy as fundamental, so every job states whether the examples contain personal data.

Propose a model for the queue

Help me draft a proposal to install one model on a shared community machine. Ask me for the model name, the code license, the weights license, whether Swedish text is in scope, the disk space, and the task, which is either inference (the model produces an answer) or fine-tuning (further training on examples). Explain free redistribution in one sentence: people may pass the software on without paying a royalty, which is the first criterion of the Open Source Definition. Then write the proposal with a time limit for one queue job, the measurement I will publish (settings, test set, score, and date), and a yes or no on whether the examples contain personal data. If either license is missing, stop and name the file I need to read. Do not invent a license.

Membership agreement

The form will use this plain-language agreement.

The operator is Hrafninn Industries AB, Kungsbacka.
You may read the chapters with no account.
An account stores your name, email, and password. Email is used to confirm the address, send the starter, and reset the password.
The confirmation message tells you how to stop mail that the account does not need in order to function.
You keep the copyright in what you write. You allow the site to display it under your name.
You can ask for a copy of the account data and for deletion. IMY, the Swedish Authority for Privacy Protection, sets out the duties an organization has when it uses personal data.
If the station is bought, the same queue rules apply to every account, whether or not it has given money. Reading stays free whether or not you give money toward the machine.
Founding membership is the record of a gift toward the machine: your name on the chassis and a vote on which models are installed. If the station is bought, Hrafninn Industries AB will own it.

The confirmation mail will contain four things: a link to confirm your address, the three acts (read one feed item, ask one question, copy one prompt), the three-day starter, and the line that stops further mail.

Forum rules

These rules apply to the feed, to project threads, and to measurement posts.

Read the replies already there before you add one.
Put one question in a post. Include the model name and what you tried.
Leave out other people's personal data, passwords, health details, and unpublished contracts.
Say so when you write for a company. A sales pitch presented as an ordinary reader's note comes down.
Name the license when you share code or model weights.
Answer the claim. Leave insults out.
A measurement includes the settings, the test set, the score, and the date. A changed setting is a new post.

Why the chapters stay free

A person in a first job and a specialist open the same chapter. If the fund opens, gifts and company sponsorship would pay for the station. They give no share, no claim and no reserved time; founding members get a name on the chassis and a vote. The price of a chapter stays at zero.

The Open Source Definition requires free redistribution as a condition of open source software. Mozilla's manifesto says free and open source software promotes the internet as a public resource (principle 7), and that transparent community-based processes promote participation, accountability, and trust (principle 8). Principle 4 says an individual's security and privacy on the internet are fundamental. A model proposal has to pass those tests in public: a license a reader can open, a vote a reader can see, and a straight statement about personal data in the examples.

AI Sweden's technology infrastructure shows one existing Swedish pattern. Partner organizations can test hardware and datasets there, and the center says some activities may come at a cost. Polaris is raising money for a shared station, and the guides stay open to read.

If the station is bought, every account can join one queue for inference and fine-tuning of openly licensed models on a machine in Sweden. Founding members then vote on which models are installed. About the shared station

Sources

Elements of AIFree courses from MinnaLearn and the University of Helsinki. The site states the aim: what AI is, and what it can and cannot do.
Elements of AI, Swedish courseSwedish entry point for the free Elements of AI course.
Practical Deep Learning for CodersFree fast.ai course for people with some coding experience. Part 1 includes deployment, language, and a data-ethics lesson.
Hugging Face LLM course, chapter 1Free course, no ads, expects Python. Hugging Face releases the text under the Apache 2.0 license: credit them, link the license, and mark changes if you reuse it.
The Open Source DefinitionThe Open Source Initiative's criteria for open source software. The first criterion is free redistribution. The Initiative's Open Source AI section covers models.
The Mozilla ManifestoPrinciples 4, 7, and 8: privacy, free and open source software, and transparent community processes.
AI Sweden technology infrastructureHow AI Sweden, the national center for applied AI, gives partner organizations access to a testbed, datasets, and use cases. The page says some activities may come at a cost.
IMY: data protection for organizationsThe Swedish Authority for Privacy Protection on the duties that apply when an organization handles personal data, including the legal basis and security.