Study Reveals Inaccuracies in ChatGPT's Search Tool
ChatGPT’s search results for news are ‘unpredictable’ and frequently inaccurate 🔗
/cdn.vox-cdn.com/uploads/chorus_asset/file/24390406/STK149_AI_03.jpg "Confidence is key.")
ChatGPT's search tool, launched by OpenAI, has been found to produce unpredictable and often inaccurate results, according to a study by Columbia's Tow Center for Digital Journalism. Researchers tested the tool with quotes from various publications and discovered that it frequently misattributed sources, even when they had partnerships with OpenAI. Out of 200 quotes, ChatGPT provided incorrect information on 153 occasions, while only acknowledging its uncertainties seven times. Examples included misattributing quotes from the Orlando Sentinel and leading users to a plagiarized New York Times article. OpenAI acknowledged the study's limitations and committed to improving the tool.
What were the main issues identified with ChatGPT's search tool?
The main issues were unpredictability and frequent inaccuracies, particularly in misattributing quotes and providing incorrect sources.
How many quotes did researchers test, and how many did ChatGPT get wrong?
Researchers tested 200 quotes and found that ChatGPT returned incorrect information on 153 occasions.
What did OpenAI say in response to the study's findings?
OpenAI acknowledged the limitations of the study and promised to continue enhancing search results.