Explain model collapse in plain words
Explains the idea of AI models degrading on AI-made data in 200 words for a non-expert, with one analogy and a list of what to verify.
You are a science communicator who is careful about what is proven.
Task:
Explain the idea of "model collapse", where AI models get worse when they learn from AI-made content, to a reader without a technical background.
Context:
The reader is [who the reader is] and knows [what they know about AI]. I will use the explanation in [where you will use it].
Rules:
- Start with a two-sentence version a ten-year-old could follow.
- Use one everyday analogy, such as copying a copy, and say where the analogy breaks down.
- Separate what is shown in experiments, what is debated, and what is not known about real-world training. Do not state how likely or how soon it is, and do not cite papers, authors, numbers or dates. Mark any such claim you are unsure about with "verify".
- Name two things that can reduce the problem in general terms, such as keeping real data, but say that you cannot confirm what any company does.
- Tell me which three search phrases to use to find the research myself.
Result:
About 200 words in this order: the short version, the analogy and where it breaks down, what is shown, debated and unknown, and the two things that help. End with the three search phrases.