Skip to main content


😲 Kyle Kabasares, a Physics PhD graduate working at NASA's Ames Research Center, gave the methods section of his research paper to ChatGPT O1 Preview and asked it to generate the code based on the description. After just six prompts, it produced a working version of the code that took him a year to develop during his PhD. #ChatGPT #LLM #ML #AI youtube.com/watch?v=M9YOO7N5jF…
in reply to Chi Kim

All I can say is "WTF?"! Oh no... That's going too far... LOL
in reply to victor tsaran

@vick21 He has PhD in Physics. I have no idea if he has extensive background in programming, or how complex his code was. I guess the point is that if you know how to ask with the right prompt and quality check the output from LLMs, someone like him doesn't have to spend a year on coding and instead focussing on something else. Having said that, I have o1-preview, and it's great! The main drawback is you only get 30 messages per week. :)
in reply to Chi Kim

Do we know for sure that OpenAI was not already trained on his code? Curious then, if the machine would be able to produce that same code without having ever seen it? :)
in reply to victor tsaran

@vick21 He raised that question in the video. He said he migrated his code to Github, but it's a private repo. Maybe Microsoft handed over private repos to OpenAI too? lol Even if the model had his code as part of training data, he was very impressed with the output.
in reply to Chi Kim

I heard about One's capabilities from coWorkers. They are quite impressed. I don't have Plus subscription though...