Hello,
I have encountered a systematic issue with ChatGPT, where it ignores user instructions regarding information retrieval. Specifically, the bot does not follow predefined search parameters and does not always rely on evidence-based sources. This reduces the quality of the provided responses and results in a negative user experience.
Main Issues:
Failure to Follow Search Instructions
The chat settings clearly state: “In this chat, do not use information search in the Russian internet, but only in the English-language internet.”
However, despite this requirement, the bot regularly retrieves information from Russian-language sources, which contradicts the specified instructions.
Example: I ask about scientific research on trichology, expecting references to PubMed, Cochrane, NIH. Instead, the bot provides links to Russian-language websites, which does not meet my request.
Ignoring Previously Defined Rules
Even when I explicitly define rules at the beginning of a conversation, the bot starts violating them after some time.
Example: I specify that I need only evidence-based medicine, but later the bot begins to provide generalized opinions based on marketing claims from manufacturers.
Justifying Errors Instead of Fixing Them
After corrections, the bot often explains why it acted that way but does not adjust its behavior in subsequent responses.
This makes interactions unpredictable and reduces trust in the system.
Lack of Strict Evidence-Based Approach
Occasionally, the bot provides generalized advice without actual references to scientific studies.
Example: In a discussion about hair loss, the bot might mention herbal extracts (such as nettle or burdock) without citing scientific evidence of their effectiveness.
Furthermore, instead of providing rigorous evidence-based data, the bot sometimes refers to folk medicine or traditional remedies, which lack scientific validation.
When asked to support such claims with reliable sources, the bot fails to find actual studies, confirming that such responses are not only useless but dangerously close to misinformation.
The expectation is that information should be based exclusively on scientific data (PubMed, Cochrane, ScienceDirect, NCBI), but this is not always the case.
Expectations and Requests for Developers:
Improve the algorithm to ensure it strictly follows user instructions without deviations.
Guarantee that searches are conducted only in the English-language internet, if specified in the settings.
Enhance the instruction-processing mechanism so that the bot does not forget user-defined rules within a single conversation.
Implement an automated verification system for evidence-based sources to eliminate unreliable or irrelevant references.
Prevent the bot from referencing folk medicine or traditional remedies unless they are backed by scientific studies, ensuring all health-related responses meet evidence-based standards.
Please forward this request to the technical team for issue resolution.
Thank you.