My prediction: this is because of tool use. All models by OpenAI hallucinate more once tool use is given. I noticed this even with 4o with web search. With and

My prediction: this is because of tool use. All models by OpenAI hallucinate more once tool use is given. I noticed this even with 4o with web search. With and without websearch I have noticed a huge difference in understanding capabilities. I predict that O3 will hallucinate less if you ask it not to use any tools.