From 1 February to 25 June 2027. Online. Live sessions every week, 90 minutes each, plus a possible week in Valencia in late May or early June (see below).

The application asks which of four slots you could make most weeks: Tuesday or Thursday, around 9:00 or around 17:00 Central European Time.

We pick the most popular ones so that everyone - no matter which part of the world they’re joining from - can attend at least one live each week.

12 to 16 hours a week. That's 2 to 3 hours of readings and recordings before the week, about 3 hours of live sessions, 3 to 4 hours on the module activity, and from March 4 to 6 hours on your capstone.

There are buffer weeks here and there, for you to rest or catch up.

Once the modules end, you’ll have some weeks to focus exclusively on the capstone.

This is a good moment to check your schedule, if your honest answer is "I can do 8 hours a week", it’s not enough time.

Yes, you are required to attend at least one live session each week.

For the ones you miss, watch the recording and do the exercise.

Four or more missed weeks overall means we'll withdraw you from the programme.

If you already know you can't make it live most weeks, apply to another edition.

I. Foundations (Modules 1 to 4, February)

II. Evaluation Methods (5 to 8, March to mid-April)

III. Impact (9 to 12, late April to mid-May).

One module a week, a buffer week after each part.

The full diploma needs all three Parts plus the capstone.

Readings and short recordings before. Two live sessions, each a mix of teaching, discussion and/or an exercise. One graded activity after, due within two weeks. Your Slack community throughout.

Tell us early and we'll work something out. Illness, emergencies, a crunch at work: we're humans about it. What we can't do is carry someone who's gone quiet. If you stop attending and stop handing things in without talking to us, we'll ask whether this is still the right time for you.

Everything is marked out of 10 and you need an average of 5 to complete.

Module activities are about 50% of your final grade, the hands-on modules 15%, and the capstone 35%. You get a grade transcript with your diploma.

Yes, Modules 2 and 3 (architectures, metrics and experimental methodology), if you can prove knowledge. We’ll approve on a case-by-case basis.

You won’t have to pay. Your tuition is covered by a scholarship funded by Coefficient Giving.

There's no stipend, so you won't receive money for living costs or lost income.

Graduates earn an Expert Diploma, certified by ValgrAI, with a grade transcript.

The workload is equivalent to 15 ECTS, though it is not transferable as university credits.

We hope so, and we're working on the funding!

If it's confirmed, it's one week in Valencia, Spain, in late May or early June 2027.

Three things can happen: everyone goes; some teams go, chosen by the April capstone checkpoint grade, and the rest present online the same week; or it all runs online.

We'll tell you as soon as we know.

Maybe, depending on your passport. If you do, we provide an official invitation letter and guidance. You pay the visa fee and travel insurance. Schengen applications can take 4 to 8 weeks, so tell us early.

About a third of last year's cohort needed one, and two of them were not able to get one.

People who want to evaluate AI systems for a living, or already do and want to do it properly.

Graduate or PhD students, early-career researchers, and professionals with expertise in AI/ML, from academic training or professional experience.

We are mostly interested in your motivation to work on AI evaluation and safety.

This programme is particularly helpful for technical researchers in AI safety teams, or people who may occupy decision-making positions and want to have a solid understanding of all AI evaluation techniques.

Exceptional candidates from policy, governance, or social sciences with technical interests are also welcome.

You should be comfortable with code and with models at a basic level: run a notebook, call an API, read a paper and understand its method, intro statistics.

You don't need to be a research engineer. We ask what you've done hands-on in the last year, with no minimum, so we can send the right prep and balance teams.

Policy people with little coding skills have done well here; they lean on AI or teammates for the implementation and carry the analysis.

If you want a systematic view across methods you don't normally use, and to work with people from other fields on a real problem, yes.

You can skip Modules 2 and 3 if placement shows you know them, and lead the discussion in the rest.

If you want a deep specialisation in one method, there are better places.

Yes, professionals and master’s students with relevant experience are encouraged to apply.

No. We expect applicants to have technical competence in AI/ML, but the programme is designed to provide structured training in evaluation methods.

Most of last year's cohort did.

It's doable if you protect the hours and your employer knows. It isn't doable if you're also finishing a thesis and moving house. It’s a lot.

The application asks which session slots you can make, most weeks.

For our first year we received +670 applications for 40 spots.

This year we expect high interest, so being clear on your essays will strengthen your application.

The programme is conducted fully in English. You should feel comfortable following lectures, participating in discussions, and writing short assignments in English. Formal test scores are not required, but this is the level we recommend:

• The Council of Europe's CEFR:

  Recommended Minimum: B2 - Upper intermediate

• TOEFL (Test of English as a Foreign Language): Recommended Minimum: 90

• IELTS (International English Language Testing System): Recommended Minimum: 6.5

• Duolingo: Recommended Minimum: 110