Start here
New to AI? Start here
You do not need to work in tech to follow AI news. These are the few words almost every story uses, in plain English, and how to read a story on Herald of AI.
How to read a story here
- Start with the box "In short" at the top: what happened, why it matters to you, and the technical words explained. If that is all you have time for, you have the news.
- The full story follows for anyone who wants the details: nothing is left out.
- The small numbers like [1] link to the source of that fact, so you can check it yourself.
- Most stories also have a 30-second video on our Instagram and TikTok; the latest are on the front page.
The sections
- New models: the new AIs and updates to the ones you already use.
- How it works: research and new techniques, explained without the maths.
- Made with AI: films, games, art and everything people create with AI.
- People: who builds AI and what they say about it.
- The week: the most important stories of the week in one page.
The words you will meet
- Artificial intelligence (AI)
- Programs that learn from examples instead of following rules a person wrote one by one. Shown millions of photos of cats, a program learns to recognise a cat; shown a huge amount of text, it learns to write.
- Chatbot
- An AI you talk to by typing or speaking, like ChatGPT, Gemini or Claude. You ask, it answers in words, pictures or code.
- Model
- The AI itself: the "brain" behind a chatbot or an app. Companies release new models the way phone makers release new phones, and each new one is usually better at some things.
- Training
- How a model learns: it is shown enormous amounts of text, pictures or sound and adjusts itself to get better at predicting them. It takes thousands of powerful chips (GPUs) and can cost millions.
- Parameters
- The numbers inside a model that training adjusts. "70 billion parameters" means a bigger model: often smarter, but slower and more expensive to run.
- Tokens
- The small pieces of words a model reads and writes. Prices and limits are counted in tokens; in English a token is roughly three quarters of a word.
- Open-weight or open source
- A model anyone can download and run on their own computer. A closed model can only be used through the company’s app or website.
- Agent
- An AI that does not just answer but acts: it searches the web, clicks, fills in forms or writes and runs code, step after step, to finish a task.
- Benchmark
- A standard test used to compare models, like an exam. A high score shows a model is good at that test, not that it is good at everything.
- Hallucination
- When an AI says something false as if it were true. It is why every story here links to the sources behind each fact.
- Inference
- Using a trained model: each time you ask a chatbot something, that is inference. Making it cheaper and faster is what many stories about chips and software are about.
- Fine-tuning
- Teaching an existing model a particular skill or style with extra examples, instead of training a new one from scratch.
Any other word is explained in the story where it appears.