What happens in an Ulern session
From the first calibration question to the final check: the full path of one learning session, and the research behind each step.
Aliakbar Abbasi is the founder of Ulern, a learning system where a personal AI tutor plans, teaches, and adapts for each person — and leaves evidence of what they have actually learned. He started it on a simple conviction: learning should be personal and hands-on, and every minute of it should move you forward rather than in circles.
He has spent 17 years as a software engineer and built Ulern end to end, which shapes how he writes here: not AI as a black box, but knowledge you can inspect, decisions you can trace, and readiness you can prove. He also created Graflow, an open-source framework for stateful AI workflows in Django.
His current focus is the AI literacy the EU AI Act (Article 4) now expects of organizations — turning a policy PDF into role-based learning with per-person evidence. He writes on how people learn, on using AI responsibly at work, and on building AI products that hold up. He is based in Amstelveen, in the Netherlands.
Posts by Aliakbar Abbasi
AI tutor, course, or ChatGPT: what the evidence says
Tutoring research spans 40 years, from Bloom's two-sigma finding to last year's AI-tutor trials. What holds up, what doesn't, and what to look for.
How to compare learning methods fairly
Flashcards, apps, lectures, tutors: each counts learning differently. A three-question framework that puts any method on one scale.
Most tests never get past memorization
A study of 9,713 exam questions found 93% only asked students to remember or restate. What the levels above memorization are, and how to test for them.
How many times should you practice one thing?
Memory research has a surprisingly specific answer: three successful retrievals, then stop. Where the number comes from, and what makes it last.
Why learning that works feels like it isn't working
Randomized studies find the methods that teach you most feel the least productive in the moment. The evidence, and how to stop grading by feel.
Why 4.8-star learning apps get 1-star reviews
We mined the negative reviews of 12 learning apps. The dominant complaint is not learning, and the churn pattern is the same everywhere.
Introducing AI with your works council, not past it
The AI Act's worker-information duty now applies from December 2027. Codetermination rights apply today. How to introduce AI tools with the council.
Your agencies are inside your Article 4 scope
Article 4 covers your staff and other persons using AI on your behalf — agencies, freelancers, outsourced support. What that means in practice.
Bring-your-own-AI is a GDPR problem first
A personal AI account at work means no processing agreement and training by default — and a US court order showed 'deleted' chats can persist.
Article 4 was rewritten — 'ensure' became 'support'
The Digital Omnibus softened Article 4 from ensuring a sufficient level of AI literacy to supporting its development. What changes for your program.
The EU AI Act on 2 August 2026 — what applies now
High-risk duties moved to 2027 and 2028, but 2 August 2026 still activates transparency rules and AI literacy supervision. The post-Omnibus timeline.
Five AI literacy claims that are no longer true
No €35M fine for skipping training, no mandatory certificates, no guaranteed level. What the AI Act actually asks after the June 2026 amendments.
AI literacy evidence — prove effort, not scores
The amended Article 4 asks for measures, not test scores — so the record of your measures is the compliance artifact. What that record should contain.
What Article 4 of the EU AI Act asks of your team
Article 4 has applied since 2 February 2025 and was rewritten in June 2026. What the AI literacy duty asks in practice — by role — and how to evidence it.