gpt4o发布会

GPT-4o has safety built-in by design across modalities,through techniques such as filtering training data and refining the model’s behavior through post-training.We have also created new safety systems to provide guardrails on voice outputs.GPT-4o通过过滤训练数据和通过训练后改进模型行为等技术，在设计上具有跨模态的内置安全性。我们还创建了新的安全系统，为语音输出提供护栏。We’ve evaluated GPT-4o according to our我们根据GPT-4o评估了[Preparedness Framework](https://openai.com/preparedness)and in line with our[voluntary commitments](https://openai.com/index/moving-ai-governance-forward/)。Our evaluations of cybersecurity,CBRN,persuasion,and model autonomy show that GPT-4o does not score above Medium risk in any of these categories.This assessment involved running a suite of automated and human evaluations throughout the model training process.We tested both pre-safety-mitigation and post-safety-mitigation versions of the model,using custom fine-tuning and prompts,to better elicit model capabilities.准备框架，并符合我们的自愿承诺。我们对网络安全、CBRN、说服力和模型自主性的评估表明，GPT-4o在这些类别中的任何一个类别中的得分都不超过中等风险。该评估涉及在整个模型训练过程中运行一套自动化和人工评估。我们使用自定义微调和提示测试了模型的安全缓解前和安全缓解后版本，以更好地激发模型功能。

GPT-4o（“o”代表“omni”）是迈向更自然的人机交互的一步——它接受文本、音频和图像的任意组合作为输入，并生成文本、音频和图像输出的任意组合。它可以在短短232毫秒内响应音频输入，平均为320毫秒，这与对话中的人类响应时间相似。它在英语文本和代码上的GPT-4 Turbo性能相匹配，在非英语语言的文本上也有显著改进，同时在API中也更快且便宜50%。与现有模型相比，GPT-4o在视觉和音频理解方面尤其出色。

The presenters highlighted the benefits of combining fine-tuning and RAG for optimal performance.演讲者强调了将微调和RAG相结合以获得最佳性能的好处。Application of Theory理论应用The presenters applied the theory to a practical challenge,the Spider 1.0 benchmark,achieving high accuracy using both RAG and fine-tuning.演讲者将该理论应用于实际挑战，即Spider 1.0基准测试，使用RAG和微调实现了高精度。Conclusion结论The presentation concluded with a summary of the optimization flow and emphasized the importance of iteratively improving LLM performance using the discussed techniques.演讲最后总结了优化流程，并强调了使用所讨论的技术迭代提高LLM性能的重要性。Q&AThe presenters invited questions from the audience and were available for further discussion.演讲者邀请观众提问，并可以进行进一步讨论。