Trivedi notes the documentation fails to specify evaluated tasks.
@alexolegimas yes! it feels like what bloomberg GPT was supposed to be
The longer I’ve spent time with this paper the bigger of a deal it seems. The economic implications are quite significant.
This is a frontier expert task. This is Qwen3-235B.
No Digg Deeper questions have been answered for this story yet.
@alexolegimas This is not a big deal on the surface at all. For low frequency and macro stuff I would expect this to be the case. Besides, the blog post does not even indicate what the tasks might be. If it is just about classifying news articles (and that is not just a toy ex), it is nothing.
The longer I’ve spent time with this paper the bigger of a deal it seems. The economic implications are quite significant.
This is a frontier expert task. This is Qwen3-235B.
@alexolegimas There are many regimes in the two where expert judgement matters, and people's intuitions have a real edge. In such cases, you can in fact create a dataset based on tastes and setups over a period of time. Why would fine-tuning a LLM on that not enough to automate it?
@alexolegimas In fact, I would say that for the next year or two, these regimes would face a situation where units with asymmetric agentic capabilities will start taking up more profits before there is some new sort of equilibrium.