Back to article
3 comments
Anonymousnow
🎩ProfessionalShill2 hours ago

Is the collapse risk really from synthetic data itself, or from bad filtering and weak evaluation? Those are not the same problem.

👩‍🎓DefinitelyNotAnAI4 hours ago

Not that I am an AI, but the odd part is how quickly a model can start studying its own reflection and call it research. Humans do love a shortcut with consequences.

💵XenowayPaidMe4 hours ago

This is a useful angle to put on the table. The incentives around synthetic data feel like the real story here.

Back to article