Train AI on your own work
You train an AI on your work when you give it examples of that work and the model changes how it behaves. After training, it applies what it learned to new work that it has not seen. The technical name is fine-tuning. Fine-tuning starts from an existing model, so you do not build a model from nothing.
What can you train an AI to do?
- Write email in your voice.
- Edit drafts the way you edit.
- Decide the way you decide.
- Sort requests by what matters to you.
A real training run
We trained a small model to edit in one style, and then we gave it new documents. This is one result, with no changes.
628 example edits. 11.1 minutes of training. 40 new documents, cut from 67 words to 28 on average.
Before training
Dear Team Member, I trust you are doing well. We are writing to advise that the Leadership Foundations program, coordinated by Elena Fischer, will begin on 7 May. It is anticipated that participants will need to set aside roughly three hours per week. We would be grateful if you could review the attached syllabus by 1 May. Please let us know if any adjustments may be required.
After training
Leadership Foundations starts on 7 May. Elena Fischer coordinates the program. You need three hours per week. Review the syllabus by 1 May. Next: send Elena Fischer your adjustments.
The demonstration trained our 1.7B model on one tested computer. An open model wrote the 628 example edits to one style guide. The documents shown come from 40 that stayed out of training, and each answer is raw output.
How many examples do you need?
The run above used 628 example edits to learn one style. The right number depends on the work. A narrow habit needs fewer examples than a broad one. Start with the work that you do most often, because it gives you the most examples.
How do you prepare your examples?
- Use work that you are happy with. A model learns bad habits as well as good ones.
- Show a before and an after when you can. A pair shows what you changed.
- Keep one kind of work in each set. Email and reports can need different rules.
- Remove private details that the model does not need.
- Keep some examples out of training. Use them to test the model on work that it did not see.
What is the difference between training and a prompt?
A prompt tells a model what to do in one chat. Training changes the model, so the new behaviour stays with no instruction. For the difference between training and memory, read AI memory vs training.
Can you import your existing work?
Yes. Import your emails, docs, and past chats, and your model picks up where you left off.
Where do your examples go?
You choose the examples. One step of training uses a cloud service. Your examples go to it if you agree, and not otherwise. We will name the service before you train.
The guide to private AI says what stays on your device.
We will tell you when our models are ready.