GPT-5.4 NANO
The GPT-5.4 nano’s answers are somehow funny, because its errors aren’t just contextual misunderstandings, but also patterns and even numerous word choices, making it seem like guesswork. Only two questions were consistently answered correctly, representing only about 10% of the data.
This results shows that while I initially thought the data I created was too easy, it turns out that it’s still quite challenging for a small model.
Hello, I’m Lusiana!
Welcome to my learning adventure!
I’m interested in learning new things and am currently interested in Artificial intelligence (AI).
The Coralab is my “imaginary” laboratory. I’ll be posting about the things I learn about Artificial intelligence (AI) here.
PS: Btw, this lab is available in dark and light mode. Enjoy!
PPS: Actually, I still can’t believe I’m back writing blog after several years. Usually when I write blog, I don’t write fiction and vice versa, but now I’m doing both, so good luck for me and my energy.
I might try out various services and write about my experiences using them. My posts may contain praise, criticism, or suggestions, but are not intended to attack or promote any particular brand.
The goal is to capture real-world user experiences. I’ll clearly define the date and model type to avoid confusion or ambiguity. (If needed)
git(at)lusianaliu.com
