The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Gemini has gotten better at shitlaw
| Dave Prole | 07/08/25 |
| https://imgur.com/a/o2g8xYK | 07/08/25 |
| ,.,.,.,....,.,..,.,.,. | 07/08/25 |
| https://imgur.com/a/o2g8xYK | 07/08/25 |
| """'"'"""'' | 07/08/25 |
| https://imgur.com/a/o2g8xYK | 07/08/25 |
| scholarship | 07/08/25 |
| Dave Prole | 07/08/25 |
| https://imgur.com/a/o2g8xYK | 07/08/25 |
| ,.,.,.,....,.,..,.,.,. | 07/08/25 |
Poast new message in this thread
Date: July 8th, 2025 7:38 PM Author: Dave Prole I painstakingly had gpt and Claude separately review a hostile client situation and draft a proposed email. Claude was more breezy and casual and GPT was robotically professional.
Had Gemini Pro 2.5 look at both drafts and pick one, it said they're both shitty and absolutely DO NOT SEND EITHER. Gave me a huge list of bulleted feedback and cited actual bar ethics / rules. I told it to just write the damn thing itself from scratch
Then I showed the Gemini draft to Claude and GPT and both were like "oh yeah that's way better than mine" |
Date: July 8th, 2025 7:39 PM Author: https://imgur.com/a/o2g8xYK If all you need is draft language, you can run Gemma on a $200 graphics card and get equally good results. It's just a quantized version of Gemini, and quantization doesn't really affect it's writing ability (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49083887) |
 |
Date: July 8th, 2025 9:06 PM Author: ,.,.,.,....,.,..,.,.,. Gemma is not anywhere close to 2.5 pro. Completely different models (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49084199) |
|
Date: July 8th, 2025 9:08 PM Author: https://imgur.com/a/o2g8xYK For writing shitlaw motions and pleadings it's fine. I'm on Gemma3 12b, but Gemma2 was good enough. And no, Google didn't train two totally different LLMs but only commercialize one of them. lmao they both got trained on Google's in-house legal docs (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49084213) |
|
Date: July 8th, 2025 7:40 PM Author: """'"'"""'' oh great, now we will have two rats suing each other with an army of AI agents. (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49083889) |
|
Date: July 8th, 2025 7:51 PM Author: https://imgur.com/a/o2g8xYK No, legal writing will just become less SHITTY than it currently is. Sorry you SUCK at it too much to appreciate the difference (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49083927) |
|
Date: July 8th, 2025 7:55 PM Author: scholarship It’s fucking awesome at it. Been paying for Claude too but it beats Claude pretty regularly and sometimes just blows everything else out of the water. Got best overall on BigLaw Bench recently. (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49083941) |
|
Date: July 8th, 2025 9:01 PM Author: Dave Prole Yeah it's awesome at shitlaw
I think it would get wrecked at anything complicated like appeals or medmal tho |
|
Date: July 8th, 2025 9:06 PM Author: https://imgur.com/a/o2g8xYK It gets antitrust up and down, if that's complicated (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49084202) |
|
Date: July 8th, 2025 9:05 PM Author: ,.,.,.,....,.,..,.,.,. The gemini deep think model is supposed to launch soon and looks significantly better. Going to be 180 (http://www.autoadmit.com/thread.php?thread_id=5747770&forum_id=2#49084198) |