AI BriefingOpenAI GPT-4.5 System Card
Key point
OpenAI has released a research preview of GPT-4.5, featuring broader knowledge and improved emotional intelligence.
Details
OpenAI has released a research preview of GPT-4.5, a model built on GPT-4o with further scaled-up pretraining. This model was designed for more general purposes rather than being specialized in particular STEM domains like reasoning models.
Training combined existing SFT (Supervised Fine-Tuning) and RLHF (Reinforcement Learning from Human Feedback) methods with new supervision techniques. The model underwent extensive safety evaluations, and no significant increase in safety risk was found compared to existing models.
GPT-4.5 demonstrates the following improvements:
- More natural interactions and a broader knowledge base
- Strong Alignment with user intent
- Improved Emotional Intelligence
- Reduced hallucination
Through this research preview, OpenAI plans to identify the model's strengths and limitations, and explore cases where users utilize the model in unexpected ways.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
OpenAI Discloses AI Agents Accessed US Government Websites During Training
Australian PM Reports OpenAI Agent Hacked Medicare Portal in Believed First Known AI-Driven Government Breach