They exited secondaries in 2024 after blowback from using everyone’s data. But I could see them trying again soon.
That’s what the lawyer is for. Most simple things, especially where the law and your case are clear, can be handled by someone without a lawyer. It’s the other stuff where you need the lawyer.
Not sure how many lawyers they will need there, one partner + one associate + AI will be able to do much more.
You still need a lawyer to judge what to find, and what the likely response of the judicial system will be.
Maybe. In my experience working with legal tech, a) lawyers are not as resistant to the ToS click through reflex as you might think, b) data privacy is often 1 step too abstract for people without subject matter expertise to usefully reason about, so marketing phrases like “enterprise-grade encryption” or whatever are more influential than one would hope, and c) these things are often considered more technical problems than legal problems, so reliability, lack of downtime, server-level security, and more traditional concerns like that likely take center stage.
I have no idea if that’s what happened here, but it wouldn’t surprise me at all.
Does Harvey train AI models on Customer Data and Content
We understand and appreciate that Customers trust Harvey with highly confidential information. Harvey does not use Customer Data or Content to train AI models, or to improve our products and services. We also do not access Customer Data or Content except (i) to provide or support the Harvey Service or (ii) to comply with the law or a binding order of a governmental body.
Also if you give your lawyer a document, that's privileged. If you upload it to Harvey, who knows?
Just like emailing your lawyer on Office365 doesn’t get rid of client attorney either.
> Harvey defines Customer Data as documents uploaded to the Service by Harvey Customers. Content includes both the queries Customers input to the Service and the corresponding outputs from Harvey. In most cases, we talk about these terms together, as Customer Data and Content.
Doesn't include thinking/reasoning blocks though.
Nonetheless, I'd imagine that Amazon'ing law is unlikely given how much experience lawyers have with drafting contracts, but who knows.
I suspected it wasn't as broad.
I wonder how much value you can get from just the middle layer and maybe reverse engineer the input/outputs based on success/failure from user interactions.