

820: OpenAI's o1 "Strawberry" Models
25 snips Sep 20, 2024
Explore the groundbreaking capabilities of OpenAI's latest o1 'Strawberry' models. Discover how these models revolutionize AI with advanced reasoning skills, mirroring human thought processes. Delve into their strengths and limitations as they signify a potential turning point in generative AI technology. Gain insight into the future implications of these models, especially in relation to the concept of singularity.
AI Snips
Chapters
Books
Transcript
Episode notes
Scaling Inference Time
- O1's performance improves with increased "thinking" time, unlike previous models.
- This allows for scaling through inference time, potentially leading to breakthroughs with longer durations.
Choosing the Right LLM
- Use O1 for complex tasks needing deliberation, like coding or math problems.
- For simpler tasks like email or editing, other LLMs like GPT-4 are still suitable.
Counting 'r's in 'strawberry'
- O1 can count the 'r's in 'strawberry', demonstrating improved language processing.
- Previous models struggled with this due to word tokenization into subwords.