# Renaissance Labs ยท Learn > Films and lessons on how language models are trained and how human minds work. Free, sourced, in plain language. ## Course - [What a language model actually is](https://learn.renaissancelabs.eu/lessons/what-is-a-language-model/): A language model is a function that predicts the next piece of text. Everything else, from chat to code, is built on top of that one trick. - [Tokens: how text becomes numbers](https://learn.renaissancelabs.eu/lessons/tokens/): Models read and write tokens, not words. Tokens decide what you pay for, how much fits in context, and why models stumble on spelling and arithmetic. - [Attention and the transformer](https://learn.renaissancelabs.eu/lessons/attention-and-the-transformer/): Attention lets each token look back at earlier tokens and decide which ones matter. Stacked many times, it is the core of every modern language model. - [How a model learns](https://learn.renaissancelabs.eu/lessons/how-a-model-learns/): Training is a loop that measures how surprised the model was by real text and nudges its parameters to be less surprised next time. - [Pretraining data](https://learn.renaissancelabs.eu/lessons/pretraining-data/): Frontier models read tens of trillions of tokens, but most of the raw web is thrown away first. Quality, mixture and repetition shape what the model learns. - [Scaling laws: how big, how long](https://learn.renaissancelabs.eu/lessons/scaling-laws/): Loss falls smoothly as parameters, data and compute grow. Scaling laws turn that into a budgeting tool, and also show where the tool stops working. - [Parameters, precision and memory](https://learn.renaissancelabs.eu/lessons/parameters-precision-and-memory/): A parameter is a stored number, and every number costs bytes. Precision, gradients and optimizer state decide whether a model fits on one chip or needs a cluster. - [GPUs and parallelism](https://learn.renaissancelabs.eu/lessons/gpus-and-parallelism/): A GPU is fast only when its arithmetic units stay fed. Training a large model means splitting it across thousands of GPUs in several ways at once, and paying for the communication. - [Mixture of experts](https://learn.renaissancelabs.eu/lessons/mixture-of-experts/): A mixture-of-experts model stores many specialist sub-networks but sends each token to only a few. It gets the capacity of a huge model at the arithmetic cost of a much smaller one. - [When training breaks](https://learn.renaissancelabs.eu/lessons/when-training-breaks/): At frontier scale something is always failing. How labs survive broken chips, loss spikes, silent wrong answers and fragile low-precision numbers. - [Supervised fine-tuning](https://learn.renaissancelabs.eu/lessons/supervised-fine-tuning/): How a raw text predictor becomes an assistant by studying examples of good answers, and why a small, clean dataset can beat a large, messy one. - [Preference tuning and reinforcement learning](https://learn.renaissancelabs.eu/lessons/preference-tuning-and-rl/): How models learn from comparisons and from checkable rewards: RLHF, DPO, verifiable rewards, group-based RL, and the sandboxes and reward hacking that come with it. - [Evaluation](https://learn.renaissancelabs.eu/lessons/evaluation/): How we measure what a model can do, and why every measurement has a catch: contamination, saturation, judge bias and scores that cannot be compared across versions. - [Distillation and quantization](https://learn.renaissancelabs.eu/lessons/distillation-and-quantization/): Two ways to make a model cheaper to run: teach a small model to imitate a big one, or store the same weights in fewer bits. What each one keeps, and what it costs. - [Inference and sampling](https://learn.renaissancelabs.eu/lessons/inference-and-sampling/): How a trained model produces text: a fast reading phase, a slow one-token-at-a-time writing phase, a memory called the KV cache, and the sampling rules that pick each token. - [What "open" means for AI models](https://learn.renaissancelabs.eu/lessons/what-open-means/): "Open" can mean downloadable weights, public code, released data or a full training recipe. These are different things, with different freedoms, and few releases include all four. - [How a base model is trained, end to end](https://learn.renaissancelabs.eu/lessons/base-training/): Every step from raw text to a base model, followed through one real model, the OpenEuroLLM 9B that sits under Agora, with interactive demos of the loss, the optimizer, the schedule and long context. - [How a base model becomes an assistant, end to end](https://learn.renaissancelabs.eu/lessons/post-training/): Seven stages of post-training followed through one real model, Agora 9B, with its data, recipes, costs and scores, and interactive demos of loss masking, preference tuning, weight merging and reinforcement learning. ## Notes - [In practice there is a difference](https://learn.renaissancelabs.eu/notes/theory-and-practice/): Running a job on a cluster is a skill of its own. Making something work before you understand why is a legitimate first step, as long as you come back for the why. ## Films - [Keeping AI agents safe: injection, incidents and design](https://learn.renaissancelabs.eu/films/agent-safety/): Why agents raise the stakes over chatbots, how prompt injection works, real incidents, and the design rules and habits that limit the damage. - [Post-training a European model, stage by stage](https://learn.renaissancelabs.eu/films/post-training/): A walkthrough of turning a raw 9-billion-parameter European base model into an assistant, with the bugs, evaluations and fixes found at each stage. - [Envy is a compass](https://learn.renaissancelabs.eu/films/envy/): Envy is not proof that you are a bad friend. It is precise, and it points at what you still want. - [The mind is a machine that makes up stories](https://learn.renaissancelabs.eu/films/mind-making-up-stories/): The mind turns what happens into stories about you and the world. Seeing them as stories makes room for other, more accurate ones. - [What is intelligence?](https://learn.renaissancelabs.eu/films/intelligence/): A short essay in film: intelligence as finding and keeping the right pattern, and how brains and language models share that shape. - [The alignment problem: whose values should AI follow?](https://learn.renaissancelabs.eu/films/alignment-problem/): How assistants learn behavior from human choices, four questions where reasonable people disagree, and the ways ordinary people can take part. - [Why AI makes things up and how to guard against it](https://learn.renaissancelabs.eu/films/hallucinations/): Why language models state invented facts with confidence, how a model can learn to say it does not know, and what you can do to protect yourself. - [Where the silicon ends](https://learn.renaissancelabs.eu/films/gpu-hardware/): Why Hopper cannot run 4-bit math at full speed and Blackwell can, and which GPU limits are carved in silicon rather than code. - [How AI agents work: build one from scratch](https://learn.renaissancelabs.eu/films/how-agents-work/): Builds an agent piece by piece: a model that predicts text, a context list, tools, a loop, then memory, skills, sub-agents and autonomy. - [Memory is an experience](https://learn.renaissancelabs.eu/films/memory/): Each time you remember, you rebuild the memory in the present, which means the way you see the past can change. - [Running on empty: understanding burnout](https://learn.renaissancelabs.eu/films/burnout/): If you wake up already tired, you may not be lazy. Burnout has a shape that research describes, and five steps that help you come back. - [How reinforcement learning scales for language models](https://learn.renaissancelabs.eu/films/rl-scaling/): How labs run reinforcement learning at production scale: the algorithms, the environments and rewards, the infrastructure and the failure modes. - [Looking inside AI: the science of interpretability](https://learn.renaissancelabs.eu/films/interpretability/): How researchers look inside neural networks: superposition, sparse autoencoders, features, circuit tracing, and what this may mean for safety. - [Understanding your attachment style](https://learn.renaissancelabs.eu/films/attachment-heal/): Why you spiral, withdraw or push people away, how each pattern grew, and how to practice something steadier and support someone you love. - [What it takes to train a frontier model](https://learn.renaissancelabs.eu/films/frontier-training/): A tour of what goes into training a 2026 frontier model: data, scaling, chips, parallelism, failures, post-training and cost. - [The only place experience happens](https://learn.renaissancelabs.eu/films/present/): What the present moment is, why minds wander, and what the evidence says about mindfulness and awe. - [Inkling 2 Small: training in 4-bit](https://learn.renaissancelabs.eu/films/inkling/): A look at how Inkling 2 Small was trained in native 4-bit, what the run probably cost, and how a smaller team could climb to the same recipe. - [You are not alone in here](https://learn.renaissancelabs.eu/films/not-alone/): We see our own inside and everyone else's outside. This film looks at why that makes us feel separate, and at small ways to feel connected again. - [Hello, old friend: living with fear](https://learn.renaissancelabs.eu/films/old-friend/): Fear is an old companion trying to keep you safe. A gentle film about neither obeying it nor fighting it. - [How to feel and still choose](https://learn.renaissancelabs.eu/films/emotion-regulation/): Emotional regulation is not feeling less. It is a skill, built mostly with other people, and five practices can strengthen it. - [Entropy: order and surprise](https://learn.renaissancelabs.eu/films/entropy/): Shannon's entropy, followed through physics, the entropic brain hypothesis and the training loss of a language model. - [Act first, and do what you say](https://learn.renaissancelabs.eu/films/act-first/): Motivation often follows action. A smaller, kinder way to close the gap between what you say you will do and what you do. - [A simple model of how the mind processes information](https://learn.renaissancelabs.eu/films/how-the-mind-processes-information/): A simplified model of how senses, feelings and thoughts feed one another, and why changing your inner models can change how the world feels. - [Inside the strongest open models](https://learn.renaissancelabs.eu/films/open-models/): Ten leading open-weights models compared: sparse experts, compressed attention, stability, data, and why post-training seems to decide the score. - [Running AI agents safely on a cluster](https://learn.renaissancelabs.eu/films/cluster-agents/): How to let a learning agent use the real internet without letting it do real damage: sandboxes, egress proxies, logging and staged rollouts. - [Make a Fool of Yourself](https://learn.renaissancelabs.eu/films/fool/): Why the fear of cringe feels so big, why it is often wrong, and a week of small, deliberate foolishness. - [The future is a story you tell today](https://learn.renaissancelabs.eu/films/future/): How the brain builds the future from memory, where it goes wrong, and four gentle questions that make the story kinder.