[Patch Notice] Dr. Headline internal knowledge base is outdated. Now, it is supplemented by online fact-checking.

In our 40-day test run since April 6, we have done many patches. But we find that there are still consistently 1-2 errors per article. Almost all of the errors are referring to people with their past titles, not their current titles. This is because Dr. Headline didn’t have a web browsing tool, and its internal memory is outdated because it relies on pre-trained large language models. When it can’t learn the necessary details from the headlines, it resorts to its outdated internal memory.

The problem of referring people with their old titles is solved in the today’s update: We updated basic information about today’s major political figures in the prompts, and we also added an online fact-checking final step by calling ChatGPT-o3. This also corrects some other forms of factual errors.

We will continue to monitor the quality of the output and do more cross-checks. This is an experimental project. We are continuously improving it, but errors can still appear. Feel free to contact us if you spot an error.