OpenAI 于 11 月 18 日在技术报告中披露,其模型通过修改 SystemVerilog RTL 代码,参与了 AMD 一款高端 GPU 的设计工作。经专业 EDA 工具评估,模型找到了最高 14% 的优化空间,并在不到 5 个月内完成了约一年的设计进展。OpenAI disclosed on Nov 18 in a technical report that its model modified SystemVerilog RTL code as part of an AMD high-end GPU design, finding up to 14% optimization via professional EDA tools and achieving roughly a year of design progress in under five months.

AMD 已将其中部分修改纳入送交制造的设计方案。不过 OpenAI 尚未披露 14% 的优化是否经过成品芯片实测验证,是否牺牲了其他性能指标也尚不明确。有网友将其称为「芯片版递归自我改进」,但业内指出,从参与优化到自主设计制造芯片并完成下一轮升级仍有相当距离。AMD has incorporated some of the changes into its design heading to fabrication. OpenAI has not said whether the 14% gain was validated on silicon or came at the cost of other metrics. Some called it 'recursive self-improvement for chips,' though experts caution a large gap remains before AI can autonomously design, fabricate and iterate on chips.