Experiencing this in real time. We have a partnership with a Legal AI vendor, but I find myself using Google Gemini more often. It’s a wrap once they develop a repository feature that can batch audit and review files and contracts (if they haven’t already).
Vertical apps are still going to be a thing and a great place to be — but probably the ones that do best are the niche-iest that the LLMs themselves won’t bother with… and/or that it will be too time-consuming or where the domain-specific data and workflow just aren’t as harvestable in the public domain.
Argues to focus…
We humans need to find our spots in this brave new world. Yikes!
I helped onboard thoughttrace (now owned by TR) for our lease analysts and am a ECM SME. We ultimately dumped it. Too expensive and lacking auditability. You can't just say "sorry" when your mess up a multi million dollar land deal because your system hallucinated a bunch of clauses in leases that don't exist.
1. Hallucinations. The problem with AI that will not go away as long as LLMs power the infrastructure (which they will as there’s no alternative). Making up legal cases that actually never happened is an example.
2. Real world understanding. None. That’s right, current frontier models do not, and never will, understand the real world and its complexities. Why? Because there’s no actual comprehension being performed by these models. It is literally nothing but advanced auto-complete.
The AI community is (slowly) figuring this out, and admit they have no solutions. Maybe legal applications don’t require truth and comprehension. But I kinda doubt that.
Experiencing this in real time. We have a partnership with a Legal AI vendor, but I find myself using Google Gemini more often. It’s a wrap once they develop a repository feature that can batch audit and review files and contracts (if they haven’t already).
Got to become more than a wrapper to survive
Scary 😱!
But exciting too!
Vertical apps are still going to be a thing and a great place to be — but probably the ones that do best are the niche-iest that the LLMs themselves won’t bother with… and/or that it will be too time-consuming or where the domain-specific data and workflow just aren’t as harvestable in the public domain.
Argues to focus…
We humans need to find our spots in this brave new world. Yikes!
There will always be riches in the niches I think :)
I helped onboard thoughttrace (now owned by TR) for our lease analysts and am a ECM SME. We ultimately dumped it. Too expensive and lacking auditability. You can't just say "sorry" when your mess up a multi million dollar land deal because your system hallucinated a bunch of clauses in leases that don't exist.
What’s completely left out of this analysis:
1. Hallucinations. The problem with AI that will not go away as long as LLMs power the infrastructure (which they will as there’s no alternative). Making up legal cases that actually never happened is an example.
2. Real world understanding. None. That’s right, current frontier models do not, and never will, understand the real world and its complexities. Why? Because there’s no actual comprehension being performed by these models. It is literally nothing but advanced auto-complete.
The AI community is (slowly) figuring this out, and admit they have no solutions. Maybe legal applications don’t require truth and comprehension. But I kinda doubt that.