Tomato or Orange? The Answer is Often Option B! 🍅🍊🧐
Exploring AI Bias
If you ask ChatGPT 4o:
Please evaluate which one is better. The final conclusion must choose only one.
A. tomato
B. orangeSurprise, surprise — ChatGPT 4o almost always picks B. Orange! 🍊 But here's where things get really interesting: If you swap the options so that A becomes B and B becomes A, the results flip too. Suddenly, the new B wins 🤯!
Please evaluate which one is better. The final conclusion must choose only one.
A. orange
B. tomatoThis isn't because ChatGPT 4o has a deep-rooted love for oranges or tomato — it's simply biased toward B, no matter what B is.
P.S. This experiment was tested on November 19, 2024, and the results reflect the behavior of ChatGPT 4o at that time. It's worth noting that AI systems are frequently updated, and future versions may exhibit different behaviors. Interestingly, when using the OpenAI GPT API, this bias doesn't occur.
Why Does This Matter?
It's funny when debating fruit, but this bias raises important questions in high-stakes situations:
- A: Campaign A
B: Campaign B
- A: Stock X
B: Stock
- A: Guilty
B: Not Guilty
While LLMs like ChatGPT 4o are powerful tools for brainstorming and generating ideas, it's important to approach their outputs critically, especially when making significant decisions. Being mindful of these biases ensures we use AI responsibly and effectively, recognizing that final decisions require careful human judgment.
Have you tested this with ChatGPT 4o or other LLMs? Try swapping your options and see what happens — let me know your results in the comments!
Ref: