I saw an article arguing that public discussions are foundational for training AI models. How does that work, and does posting a comment mean it will be used for training?
How much do online discussions matter for training AI?
Answers
Discussion threads can provide examples of conversational language, questions, answers, and different ways people explain things. A model can learn statistical patterns from that text during training, though it is not being taught in real time every time someone posts.
Whether a particular comment gets used depends on what data a developer collects or licenses and includes in a training set. Posting publicly doesn't by itself tell you that a specific model has seen it.
Reply to this answer
I'd be careful with the word foundational. Public discussions can be useful training material, but models may also use books, websites, code, licensed collections, and other sources. The mix varies by model, and developers don't always disclose enough to measure any one source's contribution.
Reply to this answer
There is a privacy angle here too. Even if a discussion is public, someone writing a comment may not expect it to become training data. If that bothers you, check the service's privacy settings and data policies, and avoid posting sensitive details publicly.
Those steps can reduce what you share, but they don't necessarily remove copies that have already been collected.
Reply to this answer
There's also a quality issue. Discussions are full of useful explanations, but they're also full of jokes, mistakes, arguments, and confident nonsense. Training on them doesn't automatically make a model good at telling those apart.
Reply to this answer
Add your answer
Answers are reviewed before publication.