Why Better Language Models Still Repeat Bad Web Information
Neil argues that ChatGPT inherits misinformation and low-quality material from the web it ingests. Search engines have spent years trying to distinguish reliable information, so fluent AI output does not eliminate the underlying source-quality problem.
- ChatGPT draws information from web content.
- Search results still contain misinformation after decades of quality work.
- Bad inputs can produce confident but inaccurate AI answers.
- Brand publishing may be somewhat more reliable because of fact-checking incentives.
“So if your inputs are off, your outputs are gonna be off,”
“you're gonna get misinformation as well, inaccurate, wrong, whatever it may be.”