You shouldn’t cite Wikipedia because it is not a source of information, it is a summary of other sources which are referenced.
Right, and if an LLM is citing Wikipedia 47.9% of the time, that means that it’s summarizing Wikipedia’s summary.
You shouldn’t cite Wikipedia for the same reason you shouldn’t cite a library’s book report, you should read and cite the book itself.
Exactly my point.
Throughout most of my years of higher education as well as k-12, I was told that sourcing Wikipedia was forbidden. In fact, many professors/teachers would automatically fail an assignment if they felt you were using wikipedia. The claim was that the information was often inaccurate, or changing too frequently to be reliable. This reasoning, while irritating at times, always made sense to me.
Fast forward to my professional life today. I’ve been told on a number of occasions that I should trust LLMs to give me an accurate answer. I’m told that I will “be left behind” if I don’t use ChatGPT to accomplish things faster. I’m told that my concerns of accuracy and ethics surrounding generative AI is simply “negativity.”
These tools are (abstractly) referencing random users on the internet as well as Wikipedia and treating them both as legitimate sources of information. That seems crazy to me. How can we trust a technology that just references flawed sources from our past? I know there’s ways to improve accuracy with things like RAG, but most people are hitting the LLM directly.
The culture around Generative AI should be scientific and cautious, but instead it feels like a cult with a good marketing team.
Also worth noting that:
1. AI is arguably a surveillance technology that’s built on decades of our patterns
2. The US government is increasingly authoritarian and has expressed interest in throwing neurodivergent people into labor camps
3. Large AI companies like OpenAI are signing contracts with the Department of defense
If I were a US citizen, I would be avoiding discussing my personal life with AI like the plague.