ChatGPT search function demonstration fails, OpenAI code leak sparks controversy

SearchGPT was tested immediately after its launch, but problems occurred during the official demonstration.

Within just two days of SearchGPT's release, some people gained access to the gray-scale testing. A demo shared by user Kesku showcased SearchGPT's impressive response speed, providing answers almost in real-time.

However, errors also appeared in OpenAI's official demonstration, which were pointed out by foreign media as hallucination issues.

In actual testing, SearchGPT demonstrated the following characteristics:

  1. Tends to focus strongly on search results, sometimes requiring explicit indication of user intent.

  2. Can present information in widget form, such as weather forecasts, stock charts, etc.

  3. Able to directly integrate multimedia content like videos and music into the page.

  4. Supports location-based local information queries.

  5. Cites reference sources when explaining concepts.

  6. For paywalled content, it can only provide summaries and cannot access full texts.

  7. Admits limitations rather than forcing an answer when unable to find accurate information.

Overall, SearchGPT demonstrates powerful search and information integration capabilities, but there is still room for improvement in terms of accuracy and access to paid content. Its rapid response and diverse presentation methods have left a deep impression on users.

Original link