But wow. Wow wow wow. 30 mins and it generated a masterpiece. Litigators who naysay it are either 1) poor prompt engineers (likely) 2) too cheap/poor to afford pro

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182280)

you should basically write the prompt like an email instruction to an assistant, too. give it as much context as possible. you can just ask it how to best prompt it btw. in general the best way to figure out the best way to get what you want from AI is to ask the AI

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182321)

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182353)

Hey i do it all the time because thinking version takes WAY less time.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182369)

To fully mitigate any risk of malpractice or looking like an idiot when using AI to shitlaw you need to use at least 5 different AI platforms simultaneously IME.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182419)

The REAL turning point on AI in law will be when openAI finally integrates chatgpt into your discovery file to crawl around and look for shit. (Facts win cases, not law.) I'm hoping to see this within a year. Right now you can feed it bite-sized pieces, and the results, honestly, are not very impressive (but mostly because the inputs are not very large). But when it starts being able to watch and transcribe 773 hours of video and read 1.2M pages of text and tell me where are the places where subordinates raised concerns about possible MRSA outbreaks, where do you start to see the beginning signs of an outbreak, and for each decedent, create a timeline of their symptoms and reporting -- THAT will be useful. THAT will allow me to let go some of my entitled spoiled rich-kid shitlib associates from CLS.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182371)

does it have good memory features compared to stadnard? like it will remeber al the past shit i put in well or i have to keep reminding it?

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182386)

o3 pro remembered details i posted like 40+ questions ago like butter. And integrated them fully into future responses. Not used gpt 5 for convos that long yet

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182409)

No, to be honest I'm extremely unimpressed at the model's long-term memory. I tend to delete most of my threads after I'm done with them, but I also don't expect it to remember stuff from those threads; I don't know if this practice contributes to this issue. I have no idea why this happens and I do not view it as a 'core' criticism of AI (it also doesn't really bother me that much).

The laziness is so fucking bizarre it makes me want to snap sam altmans twiggy little fag arms off. Like half of what I want you to do is give me links to all ~50 fucking PDFs, goddamnit.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182410)

Pro/research grade I'd guess has a different memory bank than the normal "thinking" gpt. Keep it in one convo and you can do miracles with pro

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182428)

I have no idea how openai coded it now but it's scary af.

Obviously I'm violating a million and one client confidentiality rules but if you dare to try it will be completely fucking shocked and convinced your career will soon be caput

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182464)

OpenAI says they don't train on user inputs, right? So why is this any different from any other tool, like WL or whatever my investigator uses?

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182496)

While it is admittedly giving a lot of info to gpt, do you know how troublesome that discovery request would be for the other party to sit and wade through without AI? Oh wait then you could just send them a discovery request for THEIR gpt history. So there's an impasse, no?

I think gpt will quickly become like westlaw or lexis w/r/t discovery or bullshit like that. Just not realistic

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182512)

a lot of people liek it

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182397)

My big AI limitation (and yes, it's a huge one; I get it) is that I'm all in on ChatGPT as a product; I no longer do anything with claude, grok, etc. It was a deliberate choice and the kind of tradeoff we all have to make. It seemed, and still seems, like chatgpt is the accepted industry leader.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182422)

You do mass torts? Yeah that might reach the end of what gpt can do. But it's still a hell of a lot better than giving to a clueless associate

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182405)

They turned this on like 2 weeks ago

I made a huge thread about it and have been using it daily to do doc reviews in the background

It went through 1,000 pages of discovery the other day while I was doing other stuff and sifted out 5 key docs for my mediation brief that would've taken me 10-20 hours to find

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182430)

o3 pro also did that starting ~3 months ago too

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182444)

I bet you could, easy

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182476)

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182961)

LOL

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183443)

I literally used it against biglaw firms and me + partner have beat them the fuck out. I never tell partner where I get these flawless briefs from--and admittedly I ALWAYS check its references and do occasionally add bits and pieces. Literally same exact shit you would so as supervising attorney for a hotshot associate. GPT 5 I literally trust more than nearly all associates. It is pure insanity

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49182438)

That quote was it’s summary of case law on the eggshell psyche.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183673)

Their muscle memory and brains are too hard wired into doing rote doc review summaries + pdf tabbing + opening MS word to construct a brief from scratch using the same basic bitch boilerplate shit everyone has seen before

Litigators who don't have all that baggage and also have a knack for sophisticated cross-model prompting with customized backend/github plug ins and built in organic learning and refinement cycles will do better

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183086)

Subject: Don’t Forget, Jose

Hey Jose,

Just a little nudge to remind you that "everyone gets a shot" — yeah, even you. Also, in my four years of dodging insurance company nonsense, your facet syndrome is hands-down the most ridiculous case I’ve ever seen. Try not to make it worse, alright?

Cheers,

[Your Name]

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183472)

wanna loan me $5mm so i can post train some legal models and architectures? ill give you 30% return

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183402)

You need to build an LLM trained on your own doc library and every past client communication and internal training module, or pay someone to do that

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183447)

It’s good but not perfect. I use it as a first cut for drafting specific clauses, to help me think through things, to get me conversant on a certain topic, etc. but I have yet to have it draft me anything that I could send out.

I’ve also used it to double-check my reading of agreements with respect to discrete questions. It’s again good but not great - it misses a lot of nuance.

(http://www.autoadmit.com/thread.php?thread_id=5762076&forum_id=2)#49183476)