Google DeepMind: AI Research Foundations
Google Skills
Last updated August 16, 2026
Google DeepMind: AI Research Foundations is a rigorous, code-heavy curriculum for learners who want to understand modern language models by building one from the ground up. Over roughly 49 hours it works through building a small language model, representing text data, designing and training neural networks, the Transformer architecture and attention, fine-tuning, and using hardware such as GPUs to speed up training. The concept material is free; the hands-on labs and skill badge run on Google Skills and need a subscription or credits. It is designed for learners proficient in Python with a technical background.
What you'll learn
- Building your own small language model and understanding the ML pipeline
- Preparing, structuring, and representing text data for language models
- Designing and training neural networks, including overfitting and underfitting
- The Transformer architecture, attention, and next-token prediction
- Fine-tuning a small language model and approaches for larger models
- Using hardware such as GPUs to accelerate training
Frequently asked questions about Google DeepMind: AI Research Foundations
Who is Google DeepMind: AI Research Foundations for?
For university students and technically grounded learners who are proficient in Python and want to understand and build modern language models from the ground up.
What are the prerequisites for Google DeepMind: AI Research Foundations?
Proficiency in Python and a technical background (e.g. computer science, math, or physics). Pitched at university students and technically grounded learners.
Do you need to code for Google DeepMind: AI Research Foundations?
Yes — Google DeepMind: AI Research Foundations assumes you can code.
Why we suggest this course
There is little else like this: a from-scratch, DeepMind-built path through the actual machinery of language models — n-grams to Transformers, attention, fine-tuning, and training acceleration — aimed at people who want depth rather than a tour. Two things worth knowing: it expects genuine Python proficiency and a technical background (it is pitched at university students and technically grounded learners); and the hands-on coding labs and skill badge require a Google Skills subscription or credits.