Just one of many such stories, and yet more lawyers keep thinking it’s a good idea to bring unverified AI into a courtroom…
Sure, use AI to generate your documents and filings … but then take the time to verify it manually! Make sure the cited cases and laws actually exist and are actually relevant. Scan it for errors or ‘AI speak’. At least fucking read it.
I have no idea how people can be so confident in a LLM that they’d use it for something so high-stakes without checking its work!
I litteraly just went through this shit about 3hrs ago. I needed to install a flange gasket for a 2.5" pipe flange hydraulic return. A.I tells me I can’t use this particular multi layer gasket type I have because I have a flat flange.
Lo and behold I find the the manufacturer data sheet. Perfectly suitable for my application.
Like it’s one thing for a.i to fail at making shit up. But it’s a hole other fuck up when it can’t even regurgitate information correctly.
It’s worse than that. The AI isn’t part of the attorney/client relationship, so anything shared with it isn’t covered by privilege and is discoverable.
I wouldn’t trust anything to be truly private in the hands of these AI companies, though – they’re always scraping training data from wherever they can get it (legality be damned), and requests from enterprise clients are extremely valuable training data. They’ll make promises about how everything stays in-house … but then your chat history gets integrated into the new public model through its training, and maybe it’s now able to reproduce your private information when asked.
That would be a massive legal dispute that would probably end up sinking them. There’s legal agreements they can’t train or use the data. Would blow reputation and be legal volcano
Just one of many such stories, and yet more lawyers keep thinking it’s a good idea to bring unverified AI into a courtroom…
Sure, use AI to generate your documents and filings … but then take the time to verify it manually! Make sure the cited cases and laws actually exist and are actually relevant. Scan it for errors or ‘AI speak’. At least fucking read it.
I have no idea how people can be so confident in a LLM that they’d use it for something so high-stakes without checking its work!
I litteraly just went through this shit about 3hrs ago. I needed to install a flange gasket for a 2.5" pipe flange hydraulic return. A.I tells me I can’t use this particular multi layer gasket type I have because I have a flat flange.
Lo and behold I find the the manufacturer data sheet. Perfectly suitable for my application.
Like it’s one thing for a.i to fail at making shit up. But it’s a hole other fuck up when it can’t even regurgitate information correctly.
Why did you try to use AI for that in the first place? Are you not part of the problem? Glad you figured it out
Isn’t the rule of thumb that you can use it for things you can verify? They were able to verify.
It gave them bad advice
Never, ever use AI for legal review for a client.
Inviting an AI into the threads removes privilege.
You could use a local AI… that’s only running on your computer… that’s specifically trained on dumps of old cases and such…
But I guess lawyers are not known for their tech savvy.
Maybe you could say their heads are in the cloud.
lol, that too. Who knows what kind of private legal information you’re freely feeding to the AI company.
It’s worse than that. The AI isn’t part of the attorney/client relationship, so anything shared with it isn’t covered by privilege and is discoverable.
Disbarment should follow after the leak soon, for violating privilege
Enterprise usage of AI tools, at least those I have seen, is entirely private
Supposedly.
I wouldn’t trust anything to be truly private in the hands of these AI companies, though – they’re always scraping training data from wherever they can get it (legality be damned), and requests from enterprise clients are extremely valuable training data. They’ll make promises about how everything stays in-house … but then your chat history gets integrated into the new public model through its training, and maybe it’s now able to reproduce your private information when asked.
That would be a massive legal dispute that would probably end up sinking them. There’s legal agreements they can’t train or use the data. Would blow reputation and be legal volcano
Are you and I seeing the same AI companies? They have 10 legal volcanoes per week … all part of ‘moving fast and breaking things’.