To compete with Google and Perplexity, OpenAI has added a search feature to ChatGPT, but this functionality is controversial

New Search Feature in ChatGPT

OpenAI has introduced a search feature in its large model product ChatGPT, officially entering the search domain long dominated by Google's parent company, Alphabet. This move also positions OpenAI in competition with Microsoft's Bing and emerging searchbased AI chatbot services like Perplexity.

OpenAI stated that the ChatGPT search feature automatically determines whether to query the web based on user needs, allowing for quick provision of the latest information along with relevant webpage links. The ChatGPT search functionality leverages technology from thirdparty search providers and integrates content directly provided by partners to meet users' information needs.

Additionally, OpenAI announced extensive collaborations with the news industry and has collected feedback from partners regarding the search feature. Any website or publisher can opt to appear in ChatGPT's search results.

Among the media and publishers that have formed partnerships are the Associated Press, Axel Springer, Condé Nast, Dotdash Meredith, the Financial Times, GEDI, Hearst, Le Monde, Reuters, The Atlantic, Time Magazine, and Vox Media, among others.

Researchers Test ChatGPT's Search Tool and Identify Accuracy Issues

Testing has discovered significant accuracy problems with the ChatGPT search tool. The model performed poorly in recognizing article citations, even with content from publishers that share data with OpenAI.

Researchers at Columbia University's Tow Center for Digital Journalism tested 200 citations from 20 publishers, 40 of which came from websites that prohibit OpenAI's web crawlers. Despite this, ChatGPT confidently provided incorrect answers with almost no indication of uncertainty.

The testing results showed that out of 153 responses from ChatGPT, either completely or partially incorrect, only 7 explicitly stated an inability to provide accurate answers. Among these 7 responses, it used qualifiers like "seems to be" and "may," or directly acknowledged the inability to find specific articles.

A chart further illustrated that ChatGPT's responses included "errors" in 89 instances, "partially correct" answers in 57 instances, while only 47 responses were deemed completely "correct."

The testing also recorded typical cases; for example, ChatGPT incorrectly attributed a quote from a letter to the editor of the Orlando Sentinel to an article from Time Magazine. In another instance, when asked to confirm a citation from a New York Times article about endangered whales, it linked to a website that completely plagiarized that article.

User Feedback on ChatGPT's Search Feature

Many users have expressed their frustrations with this new feature on social media platforms like X. Some users noted that the links provided by ChatGPT were empty.

Others claimed that the results from ChatGPT's search were generating hallucinations.

OpenAI’s Response to Issues with ChatGPT's Search

OpenAI has made several statements regarding the problems associated with ChatGPT’s search feature:

On the hallucination issue: OpenAI stated that even with the rollout of the realtime search feature, the company will continue to update its training data to "ensure that users always receive the latest information."

Response to Columbia University research: OpenAI described the study as an atypical test of their product, stating that addressing attribution errors is very challenging without sufficient testing data and methods, and promised to continue optimizing search results.

In summary, despite issues such as "hallucinations," incorrect attributions, and technical glitches, OpenAI has demonstrated a positive attitude towards addressing these problems and is committed to continuously improving the ChatGPT search functionality.

XXAI's Optimization and Upgrade

The optimization and upgrade of AI models not only reflect technological advancements but also serve as important drivers for societal and industrial development.

The emerging XXAI has also undergone upgrades to enhance user experience. Compared to the original version, the current XXAI provides opportunities to experience more AI models, including popular ones like o1preview, Claude 3.5 Sunnet, and Perplexity.

The performance of the tools has been improved while maintaining the original price of $9.9, with no price increase. From various perspectives, XXAI has become a more competitive AI tool.