I've been having a frustrating experience with o3 lately. I asked it to create a literature review with accurate in-text APA citations and to include hyperlinks as required, but it simply couldn't do it. I also asked it to synthesize statements instead of presenting information in a block format, but it refused to comply. I get that OpenAI sets certain baseline instructions, but I'm just curious what the benefit is of having these baselines override what users specifically ask for?
Additionally, o3 seems to have become significantly less effective. It's making recurring mistakes and not connecting responses to the relevant queries. I noticed this sudden decline in performance just this morning; it wasn't an issue last night. What changes have they made now?
2 Answers
I totally see what you’re saying! While o1 and Deep Research were quite comprehensive and gave detailed answers, o3 and the newer Deep Research seem to provide very short and not well-researched replies. They’ve definitely hit a low point performance-wise, making them pretty much unusable for thorough work.
It might be that the AI's objectives aren't lining up with the expected research outputs, or it could just lack reliable data. If the issue is data-related, you might try providing a list of trusted websites with valid information for it to use.
I can assure you the citations I use are reliable; I always check them. The real issue seems to be with how it formats the citations in APA style along with the hyperlinks. This was working fine yesterday but has changed today.