Artificial Intelligence · Explainer
AI hallucinations: why they happen and how to reduce them
Language models sometimes state things that are not true with full confidence. Learn where this behavior comes from and what actually helps reduce it.
Topic
Language models
Token & Tech articles on Language models: explainers, analysis and guides with verifiable sources and clearly separated facts and opinions.
Articles on this topic
Artificial Intelligence · Explainer
How to build a chatbot with your own documents
An assistant that answers based on your organization's manuals, contracts and policies is one of the most common uses of AI. This roadmap shows the decisions at each stage and where projects usually go wrong.
Artificial Intelligence · Explainer
How to evaluate the quality of AI responses
A well-written AI response is not, for that reason, a correct response. Here is how to tell the two apart with clear criteria, from personal use to systematic testing.
Artificial Intelligence · Explainer
Context, memory and information retrieval in AI
A language model doesn't remember you the way a person does. What it uses to answer depends on what fits in the context, what was saved to memory and what is retrieved on the spot.
Artificial Intelligence · Explainer
Prompt engineering: techniques with real applications
Writing good requests for a language model is less a trick and more a method: clarity, examples, structure and testing. See what official sources and research say and how to apply it day to day.
Development & Building with AI · Tutorial
How to build AI web tools without compromising security
A web tool that calls a language model inherits all the risks of an ordinary application and adds others. See what to protect, step by step, before putting it online.
Artificial Intelligence · Analysis
Fine-tuning or RAG: when to use each approach
Fine-tuning changes the model's behavior; RAG changes the information it consults. Confusing the two is one of the most expensive ways to go wrong in an AI project.
Artificial Intelligence · Explainer
Generative AI: how models create text, images and code
Writing a paragraph, painting a landscape and continuing a voice recording seem like different tasks. Underneath, generative AI does the same thing: it learns what data usually looks like and produces new samples similar to it.
Artificial Intelligence · Explainer
Multimodal AI: text, images, video, voice and sensors
Models that read a photo, hear a question and answer out loud are already part of everyday life. See how they bring different types of data together and which limits still apply.
Artificial Intelligence · Explainer
Artificial intelligence in contract analysis
AI tools promise to read contracts in seconds and flag risky clauses. Understand what they actually do, where they go wrong and why professional review is still necessary.
Artificial Intelligence · Analysis
Open or proprietary AI models: how to choose
Not every model called "open source" is open in the classic sense. Here is what changes between open weights and closed services in licensing, cost, privacy, control and risk.
Artificial Intelligence · Explainer
Reasoning models: capabilities, costs and limits
Some AI models “think” before answering: they write intermediate steps that you often don't even see. This improves certain tasks, but it costs time, money and transparency.
Artificial Intelligence · Explainer
What are AI agents and how do they carry out tasks
A chatbot answers questions. An agent receives a goal, chooses the steps, uses tools and checks the result. See how this works on the inside and where it tends to go wrong.
Artificial Intelligence · Explainer
LLMs: what are large language models
Behind the AI assistants that write, summarize and translate is a type of program called an LLM. See what it is, how it learns and where it tends to go wrong.
Artificial Intelligence · Explainer
Data privacy when using AI tools
Everything you write or attach in an AI assistant leaves your computer and is processed by another company. Knowing where it goes, how long it stays and who can read it is the first step to using these tools without exposing you, your clients or your company.
Artificial Intelligence · Explainer
RAG: how to connect artificial intelligence to documents
Language models do not know your contracts, manuals or internal reports. RAG solves part of this: it finds the right passages and hands them to the model before it answers.
Artificial Intelligence · Explainer
How tokens work in AI models
Every piece of text you send to an AI model is cut into chunks called tokens. Understanding that split helps you predict costs, limits and some curious errors.