Kudos to the
@harvey team, this is a very big deal! Frontier labs set the floor, but every as lawyer knows, that's not enough. Lawyers do the most complex knowledge work in the world.
Harvey showed that they can post-train open weight models on one set of legal data, and these can generalize to perform well on other legal tasks (e.g. our redline benchmark🤝).
We've always known that legal is the biggest surface for LLMs after coding. Making LLMs work in legal will require more custom models, specialized data and cutting edge research like this. Well done!!