Lesson#4: Wikipedia vs. AI (Part 3)


 This is the third part of a summary of an article in the New York Sunday Times Magazine (7/23/23) about how Wikipedia is being used by AI chatbots for their current Results and success.

Jesse Dodge, a computer scientist, says "I don't think there's a way to solve the AI "haluciation problem." But he goes on to say that the idea of information retreival where an AI Chatbot essentially always consults a high-quality source to fact-check the information often including precise backlinks is a positive part of the continuous improvement in the AI process.

Further, Sandhini Agarwal and Girish Sastry, both ChatGPT developers, told Gertner that they are working on a new version of ChatGPT, called Chat-4, which focuses on improving "factual content." The process "reinforces learning through human feedback" and allows the user to discern the difference between good and bad answers.

Gertner went on to summarize the following points to help illustrate where the AI Marketplace currently is at by meeting with Chris Albon, Machine Learning Manager at the Wikipedia Foundation. Albon gave Gertner a quick training session and went over the following:
  • Albon asked ChatGPT about the ill-fated submersible "Titan" owned by the OceanGate company.
  • ChatGPT initially didn't answer because at that point it didn't know what happened to it.
  • He then advised ChatGPT to research the question "What happened to OceanGate's submersible?" to Wikipedia.
  • After ChatGPT found appropriate information it read and summarized them and sent its analysis as a response for ChatGPT. There was a delay in the usual time it took to respond.
The conclusion that Gertner walked away with was that chatbots can be deceived by how a question or "prompt" is worded. Barkeep49 (the editor of "Death of Wikipedia") feels that there is a case for good outcomes in the future, even if the long term seems less certain. Wikipedia has been using artificial intelligence in its development process. It uses AI to assist new editors or Wikipedia members by guiding them through a step-by-step process of creating articles, which is a process of rules, regulations and protocols that tend to alienate new wiki volunteers.

--- End of Wikipedia vs. AI Part 3 of 4 ---


Source: New York Times Magazine Article: Moment of Truth - Can  Wikipedia help teach AI chatbots to get their facts right, Jon Gertner, 7.23.23, page 34.

Comments

Popular Posts