GPT-o1: The Best Model I've Ever Tested 🍓 I Need New Tests!

In the video, the creator enthusiastically tests OpenAI’s new GPT-01 model, which they believe is the best they’ve encountered, showcasing its impressive performance in tasks like coding a Tetris game and solving complex reasoning questions. They highlight the model’s enhanced reasoning skills and accuracy, concluding that GPT-01 significantly outperforms its predecessors, despite a minor setback with a navigation question.

In the video, the creator expresses excitement about OpenAI’s release of the new GPT-01 model, which they believe is the best they’ve ever tested. They highlight a specific question involving a marble and a cup that OpenAI used in their official announcement, noting that it closely resembles a question they previously posed in their own rubric. The creator is eager to test the new model and compare its performance to previous iterations, indicating that they may need to develop more challenging questions to assess its capabilities accurately.

The creator begins their testing by asking GPT-01 to write a Tetris game in Python. They observe that the model’s thinking time has significantly decreased from over 90 seconds in previous tests to just 35 seconds this time. Upon receiving the code, they find that it works perfectly on the first attempt, showcasing the model’s improved performance and ability to generate a fully functional game quickly and efficiently.

Next, the creator tests the model with a question about postal envelope size restrictions. They note that previous models often failed to consider the possibility of rotating the envelope to fit within the specified dimensions. However, GPT-01 successfully analyzes the dimensions and confirms that the envelope meets the postal requirements, demonstrating its enhanced reasoning skills. The creator continues to test the model with various questions, including word count and logic puzzles, and is impressed by its consistent accuracy and nuanced understanding.

One of the more complex questions posed involves a scenario with killers in a room. GPT-01 not only arrives at the correct answer but also provides a detailed explanation of its reasoning, including the implications of counting a dead killer. The creator is particularly impressed with how the model navigates the nuances of the question, showcasing its advanced reasoning capabilities. They continue to test the model with various prompts, including a moral dilemma about pushing a person to save humanity, and find that it provides thoughtful responses.

Overall, the creator concludes that GPT-01 is a phenomenal model, outperforming its predecessors in nearly every aspect. They acknowledge that it only faltered on one question related to navigation from the North Pole, which has sparked debate among users. The creator encourages viewers to share their thoughts in the comments and invites them to like and subscribe for more content, emphasizing their enthusiasm for the advancements in AI technology represented by GPT-01.