Post #2248341

Great. I just saw the below post and I am immediately doubting the sanity of the world. Wtf are these people who want the "public opinion" on a matter doing? What would possess a person to think fabricating polling data through llms of all things is a good idea? And I definitely agree with the point NYTimes brought up, predictions are different from polling. And even for prediction, you do not use a "low" accuracy llm, you use a domain specific ML model that does prediction for you. I say llms are low accuracy because they are a generalized ml model, meaning they by definition will perform worse than a model finely tuned for a specific task. Of thats just the technical stuff, the far more obvious issue is the literal fabrication of data that is happening here. You can absolutely run an ML model to predict certain outcomes, including the public sentiment on a specific thing. But that does not actually replace asking the public on their opinion for things. It'd be like checking your weather app to check the current weather, instead of looking out of your window. Except your weather app asks chatgpt for the weather. Do you trust chatgpt to know your current weather outside? Similarly do you trust chatgpt to know what your local community thinks about important issues? Edit: this is the post I saw https://tech.lgbt/@Natasha_Jay/116357885337339271 Full link of nytimes article https://www.nytimes.com/2026/04/06/opinion/ai-polling.html