Hallucination detective
Game: eight AI claims that all sound equally sure. Guess what's true and what's made up.
In short
AI sounds just as sure about facts as about fabrications. Always check names, numbers, dates and sources yourself, especially if you act on them.
All correct answers come from our fact-checked lessons. After each choice you see where it’s covered.
🔍 Hallucination detective
An AI gave this answer. What's true, and what's made up? It all sounds equally sure.
In English text, one token is about 4 characters on average.
That's OpenAI's rule of thumb. Dutch text costs more tokens. Read the lesson ›
The free Gemini app can see 1 million tokens at once.
Without a plan it's 32,000 tokens. 1 million applies to AI Pro and Ultra. Read the lesson ›
In Mata v. Avianca, lawyers were fined $5,000 for cases ChatGPT made up.
Correct, the ruling is from June 2023. Read the lesson ›
Under the AI Act, chatbots only need to say they're AI from 2028.
That obligation (Article 50) has applied since 2 August 2026. Read the lesson ›
A language model learns during your conversation and updates its knowledge.
The model is frozen after training. Memory is an app feature. Read the lesson ›
According to Anthropic, Claude Mythos 5.1 is the same model as Claude Fable 5.1, with different safeguards.
Correct. Mythos is only available to selected organisations. Read the lesson ›
Grok 4.20 was released after Grok 4.5.
The other way round: 4.20 came in March 2026, before 4.3 and 4.5. Version numbers are labels. Read the lesson ›
The Dutch Data Protection Authority warned against the DeepSeek chatbot in February 2025.
Correct: be "very careful and reticent", because data is stored in China. Read the lesson ›