Model ML 用 GPT-5.6 提升金融工作效率
How Model ML Uses GPT-5.6 Sol to Get Finance Work Done More Efficiently
In a world where we're seeing so much AI generated slop, there seems to be a big difference between slop and real work. [music] Model ML offers offtheshelf and custom agents that operate exactly as our customers do. The first thing we did was we ran the model through our internal evaluation system and frankly it blew every metric out the water. Let's use the example where a user wants to build an IC memo. They give it a brief and they point it in the direction of the information that it wants to [music] pull.
在我们看到大量AI生成垃圾内容的世界里,垃圾内容与真正的工作之间似乎存在巨大差异。[音乐] Model ML 提供现成的和定制的代理,其操作方式与我们的客户完全一致。我们做的第一件事是通过内部评估系统运行该模型,坦率地说,它在所有指标上都表现出色。让我们以用户想要构建IC备忘录为例。他们给它一个简报,并指引它获取所需信息的方向。[音乐]
What we've seen 5.6 do is that will start building out a plan of exactly what it wants the slides to be and wants the slides to look like. It will commit to its work. So both doing the research, doing any calculations, and then building out the native graphs, tables, charts, logos, and then it will do a final step where it looks at each slide from a visually appealing standpoint. And we've not seen models do this anywhere near as good.
我们看到5.6所做的是,它将开始制定一个计划,明确幻灯片应该是什么内容以及幻灯片应该是什么样子。它将致力于其工作。因此,既进行研究和计算,然后构建原生图形、表格、图表、徽标,然后它将执行最后一步,从视觉吸引力的角度检查每张幻灯片。我们还没有看到其他模型能做到这么好。
It will then give the user an editable PowerPoint [music] with all of the traceable sources. The user should be focused on the judgment. In other [music] words, refining the assumptions, sharpening the message, not just rebuilding the analysis. Earlier models could do the work of an analyst. But the user would have to clearly break down the task specifically what it wanted the output to look like. With GPT 5.6 Soul, [music] we're finding that the agent gets far closer to the final output.
然后它将给用户一个可编辑的PowerPoint [音乐],其中包含所有可追溯的来源。用户应该专注于判断。换句话说,完善假设,提炼信息,而不是仅仅重建分析。早期的模型可以完成分析师的工作。但用户必须明确分解任务,具体说明它希望输出看起来像什么。使用GPT 5.6 Soul,[音乐]我们发现代理更接近最终输出。
[music]
[音乐]
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力