Topic · Themes
Understanding AI: all our articles
All 122 Alyriaz articles on the “Understanding AI” topic: news, analysis and explainers, newest first. To stay up to date, you can get an alert as soon as a new article on this topic is published.
Running an AI on your own computer: what it changes, and what it really takes
Running an AI directly on your computer has become accessible: total privacy, no cost per query, no dependency. We explain what it concretely requires in terms of hardware, which tools to use, what to really expect, and why an open model isn't necessarily one you can run at home.
What is a "reasoning model"? These AIs that take the time to think before answering
Deep Think mode, reasoning effort, sub-agents: recent models no longer respond immediately, they think first. We explain what a reasoning model is, what actually happens during that pause, why it improves answers, and why it sends the bill through the roof.
How does an AI create an image? The answer is stranger than you think.
AI image generators don't draw in the way we understand it. They start from a chaos of random pixels and progressively denoise it until an image emerges. We explain how diffusion models work, without jargon, and what this process implies about the nature of these creations.
What is a "multimodal" AI? When an intelligence learns to see, read, and hear at the same time
Multimodal: this term refers to AIs capable of processing several types of content (text, image, sound, video) within a single model. We explain simply what this means, how it's possible, and why this capability marks an important step towards AIs closer to our own way of perceiving the world.
What is an "open source" AI? The concept that reshuffles the cards of technological power
Open model, open weights, MIT licence, closed model: these technical distinctions actually decide who controls AI and who can have it cut off. We explain simply what an open-source AI is, what it changes in practice, and why it has become a major sovereignty issue.
What is "prompt engineering"? The art of talking to AIs properly, and why it changes everything
Prompt engineering is the art of phrasing a request to get the best out of an AI. We explain why the same AI can deliver a mediocre or excellent result depending on the question, and we share the practical principles for talking to models effectively, without being technical.
What does "1.5 trillion parameters" mean? The figure we brandish without ever explaining it
Grok 4.5 and its 1.5 trillion parameters, LongCat and its 1.6 trillion: the parameter count is brandished as a mark of power. We explain what a parameter really is, why it matters, and above all why a larger number does not guarantee a better model.
How does an AI "learn"? A model's journey, from raw chaos to the polished assistant
How does a model go from a pile of raw data to an assistant capable of answering you politely? We describe the three main stages of AI training (pre-training, fine-tuning, RLHF) with a simple analogy, to understand where its strengths and biases come from.
What is an "LLM"? The engine behind ChatGPT, Claude and Gemini, explained simply
LLM, large language model: the term is at the heart of all AI news, yet rarely explained. We break down simply how this technology that predicts the next word works, why this seemingly mundane ability is enough to produce text, code, and reasoning, and where its limits lie.
Teaching an AI to be "good": the hardest problem in the entire industry
Every AI model card system talks about alignment, undesirable behaviours, deception rates. Behind this technical vocabulary lies one of the deepest problems posed by modern AI: how to ensure that a system we only partially understand actually does what we want.
An AI gave three different answers to the same scientific question. The problem wasn't the AI
An Anthropic study found that an AI model could give 106, then 15, then 5 different answers to the same scientific question. The cause was not a lack of intelligence on the model's part, but a data infrastructure never designed for AI agents. One adjustment brought accuracy back up from 16.9% to 92.8%.
What is an "AI agent"? The word everyone uses without really explaining it
The word "agent" is everywhere in AI news in 2026, but rarely explained. We break down what sets an agent apart from a simple chatbot, how it works in a loop, and why this capability changes the way we should trust it.
What is the "context window"? The working memory that decides what your AI can actually understand
The context window is the amount of information an AI can hold in mind at once. We explain this key concept simply, why it exploded in 2026 (up to 2 million tokens), and why a large context doesn't solve everything.
What if AI ends up poisoning itself?
Model collapse describes what happens when an AI trains on content generated by other AIs: a gradual impoverishment, like a photocopy of a photocopy. A recent study revives the concern. We explore the phenomenon and the philosophical question it raises: can an intelligence that feeds only on itself still learn?
What is a benchmark? The test that decides which AI is best (and why it often lies)
Terminal-Bench, GPQA, MMLU, SWE-bench... Benchmarks decide which AI dominates the market, but they are often misunderstood and sometimes misleading. Here is how they really work, and why a good score does not always mean a good model.
Stealing an AI without hacking anyone: the Alibaba case shaking up the AI world
Anthropic claims Alibaba used 25,000 fake accounts to query Claude nearly 29 million times and train its own Qwen model on the responses. Here's what distillation is, why it's nearly impossible to prevent, and why not everyone is convinced by the accusation.
Why AIs hallucinate (and why they probably never will stop)
OpenAI announces a 65% reduction in hallucinations on GPT-5. Anthropic touts Claude's factual accuracy. And yet, all regular users keep seeing their AI make things up. Why? Because hallucinations are not a minor flaw. They are at the very core of how the technology works. An accessible, no-holds-barred explanation.
SubQ: the startup that may have killed the transformer
Since 2017 and the publication of the paper "Attention is All You Need", every major AI model has been based on the transformer architecture. SubQ, launched on 5 May 2026, is the first commercial model to break with this paradigm. Here is what changes technically, and why the AI community is deeply divided over the implications.
What is a token? And why it is probably the most important concept for understanding AI
If you read about or use AI in 2026, you come across this word constantly: token. Token limits, price per token, context in tokens. And yet hardly anyone actually explains it to you. To understand what is going on inside an AI, understanding tokens is probably the best place to start. And it is actually simple, provided someone takes 5 minutes to explain it properly.
How AI works: The Fortune Teller's Paradox
People often think an AI thinks or understands what it is being told. In reality, it is a gigantic statistical guessing tool. Discover the secret.
Other topics in “Themes”
- AI agents77 articles
- Code and developers22 articles
- Images and video20 articles
- Voice and music15 articles
- Robots and self-driving cars9 articles
- Health15 articles
- School and education16 articles
- Jobs and work34 articles
- Laws and regulation59 articles
- Cybersecurity32 articles
- Safety and risks51 articles
- Privacy and data45 articles
- Disinformation and deepfakes34 articles
- Energy and climate15 articles
- Chips and data centers40 articles
- Research and science43 articles
- Money and markets53 articles
- AI prices and costs43 articles
- Geopolitics and sovereignty58 articles
- Open source and local AI44 articles
- Culture and creativity42 articles
- Everyday life32 articles
- Philosophy and ethics97 articles
- Benchmarks and comparisons30 articles