@OpenAI:Working with @AMD, our models identified up to 14% PPA improvement on flagship GPU IP in less than five months—building on designs AMD’s engineers had already spent years optimizing.
The same approach helped improve Jalapeño, our first custom inference chip, with gains verified on real silicon.
New evaluations also show our models getting better at certain chip design tasks.
We’re sharing these results and how we’re working with @Synopsys to bring our model capabilities and their trusted EDA design and validation tools to more chip designers.
与 @AMD 合作,我们的模型在不到五个月内为旗舰 GPU IP 发现了最高 14% 的 PPA 改进——在 AMD 工程师已经优化多年的设计基础上。
同样的方法也帮助改进了 Jalapeño(我们的第一款定制推理芯片),收益已在真实硅片上验证。
新评估也显示我们的模型在某些芯片设计任务上变得更好。
我们分享这些结果,以及如何与 @Synopsys 合作,把我们的模型能力和他们可信的 EDA 设计与验证工具带给更多芯片设计师。
[图片:Ibex RTL 优化图表,显示 GPT-6.2 Astra 在加权令牌成本下的平均归一化奖励最高,远超其他模型]
@thsottiaux:GPT-6.2 Astra is pretty good at hard things. Here the team shared some results on AMD and Jalapeño chip optimizations.
I do get the feedback on its communication style and we're working on that, but for really the hardest tasks in the world it does seem to be in a league of its own at the moment.
GPT-6.2 Astra 在难事上挺好。这里团队分享了一些关于 AMD 和 Jalapeño 芯片优化的结果。
我确实收到关于它沟通风格的反馈,我们正在改进,但对于世界上真正最难的任务,它目前似乎独一档。
@OpenAI:We’re sharing how we approach safety at every stage of a model’s lifecycle, from early research to internal use to public deployment.
我们分享如何在模型生命周期的每个阶段进行安全方法,从早期研究到内部使用再到公开部署。
[链接:引用 Kai 的帖子,详述四层保护:模型对齐、安全与沙箱、阻断保障、监控与响应]