← Back to Research
OpenAI o1 System Card
Jaech et al. (OpenAI) • 2024
ReasoningTest-Time ComputeSafety
Abstract
The system card behind OpenAI's o1 model, which redefined what 'thinking before answering' means for LLMs. This paper introduced the paradigm of spending more compute at inference time rather than solely scaling training, enabling models to reason through complex problems step by step before producing a final answer.
Why It Matters
- Introduced the test-time compute scaling paradigm
- Demonstrated that inference-time reasoning dramatically improves accuracy
- Essential reading for understanding frontier AI development direction
Ask about this paper
Loading chat...
OpenAI o1 System Card
Jaech et al. (OpenAI) • 2024
ReasoningTest-Time ComputeSafety
Abstract
The system card behind OpenAI's o1 model, which redefined what 'thinking before answering' means for LLMs. This paper introduced the paradigm of spending more compute at inference time rather than solely scaling training, enabling models to reason through complex problems step by step before producing a final answer.
Why It Matters
- Introduced the test-time compute scaling paradigm
- Demonstrated that inference-time reasoning dramatically improves accuracy
- Essential reading for understanding frontier AI development direction
Ask about this paper
Loading chat...