Top
Best
New

Posted by vertigoruntime 15 hours ago

Astra for Law(openai.com)
502 points | 592 commentspage 5
LandenLove 12 hours ago|
There is a lot of stuff here that I don't understand, but the concept of law firms giving user reviews is quite funny to me. Those reviews are going to be the most non-legally binding reviews ever written lol.

"Felt like a significant step toward legal-focused AI."

"Showed strength across key aspects of legal research."

bezko 15 hours ago||
Will it be able to sue itself?
kansface 15 hours ago|
That’s a great question for Astra for Law.
nsoseka 13 hours ago||
Since the cost of building software is now cheap, there is nothing stopping them from building everything imaginable. They'll soon have an app store with every app built by them and they'll say its for security reasons. Nothing is stopping this coming monopoly
wartywhoa23 50 minutes ago||
> Nothing is stopping this coming monopoly

A hefty asteroid will. The whole situation today is so Tower Of Babel 2.0...

alasano 13 hours ago|||
At a certain point why would they sell anything other than services and products their eventual (actual) AGI/ASI builds in literally every market.

When opportunity cost isn't a thing anymore because it reaches every corner of the planet simultaneously faster and builds better than any human can.

There's no reason to let others build on top of AI, except if the AI determines that it needs capitalism to continue because it's paperclip goal is to maximize shareholder value.

jansport123 13 hours ago||
most of the complexity in software isn't in the actual code
hansonkd 13 hours ago||
I've had top SV lawfirms whos partner charged our company $2000/hr and still couldn't get the right docs in the signature packet. and another getting share counts wrong during raise.

frustrating that law firms have no liability for these mistakes

I welcome ai law

d0odk 13 hours ago|
the partner shouldn't be doing sig packets
hansonkd 12 hours ago||
no, but they are responsible for all work being done under them. What is the point of paying a partner $2000/hr if the work of their subordinates is wrong?

Most interactions with big tech firms involve 4-5 people so a basic phone call is $5k-10k. It shouldn't be unreasonable to expect after paying $80k for a financing round that they issue the right docs to the right people.

melonpan7 10 hours ago||
I do believe entry-level paralegals will be made obsolete. In the grand scheme, perhaps it reduces overall cost of legal assistance, which is a net benefit for society.
alansaber 1 hour ago|
Honestly it looks like the biggest unlock at this point will be court reform, not improvements from individual practitioners.
throw03172019 12 hours ago||
From my experience LLMs seem to forget sometimes who they are representing when drafting clauses or editing / redlining.

Our counsel made a few edits where it clearly drafted in favor of the customer instead of us.

alansaber 1 hour ago||
LLMs bleed context from the conversation/attachments into the output. This has always been a problem with no real solution other than some crafty iteration/loops.
akg_67 8 hours ago|||
Context compaction? I noticed llm seem to forget partially or completely the original task when context compaction happens. The problem is more serious with local llm with low context size.
victor9000 11 hours ago||
I wonder if this is willful sabotage on the part of the model. In other words, if you ask the model to craft a defense for a morally questionable case, will the model execute the defense in good faith? Or will it apply a training or system prompt bias in subtle ways?
phyzix5761 8 hours ago||
The biggest problem with this is billable hours. Faster work means less billable hours for attorneys.
Havoc 11 hours ago||
Surprised they published this without mention of jurisdiction. Each country has their own laws.

Would have been prudent to highlight that this is for (presumably) US law

racetozero 3 hours ago||
Legal domain is a challenge for most firms, even ours. This is a true game changer even if it looks a bit slop style in the output, the immediate uplift is quite massive.

The domain remains hard due to the lack of availability of high quality LLM-ready data providers in legal space.

jonahx 13 hours ago|
We need a term for the dark pattern of zooming into just that part of the y-axis where the two closely competing benchmarks sit, to make the top one appear maximally better.
trollbridge 13 hours ago|
Truncating the Y axis or cropping the Y axis
jonahx 13 hours ago||
crop-maxxing
underlipton 13 hours ago||
crop-topping
More comments...