1
Ask HN: How would you know if an AI model has been nerfed?
If consumers are just calling a model in the cloud, what’s actually enforcing that the model is not just a dumbed down cheaper model?
What’s stopping anthropic/openai from just running a cheaper LLM model based on how difficult the question is?