I'm like legal Jev pilled after using Unsloth's train-your-own-Jev model notebook. I made a legal Jev to classify trademarks under the Abercrombie spectrum. Training took about 7 minutes on a single Google Colab G4 lol.
Results on LegalBench's Abercrombie test:
📊 Base model Qwen3.5-4b: 40.0%
📊 My earlier GRPO fine-tune: 68.4%
📊 Unsloth custom Jev model: 75.8%
Trained model and dataset: huggingface.co/DoodDood/aber…
You can now train your own Decision model with our free notebook! 💡
Qwen3.5-4B will generate decisions instead of text on just 8GB VRAM locally.
Learn to data prep (state, questions, gold answers), train, serve.
Notebook: colab.research.google.com/gi…
Guide: unsloth.ai/docs/basics/train…
props to @flatspotrecords for this comp - the cover is a perfect throwback to the punk-o-rama epitaph compilation albums. i see this cover and start hearing pulley or rancid’s “1998” playing in my head dude
noone:
me: performatively reading @natolambert’s new RLHF book in union station
Ian retweeted
Launching the Public AI Constitution: A collaborative, crowdsourced AI constitution. Along with our best shot at v1.0, we also share failures of current frontier models.
The most severe breaches range from patterns of race science and denial of documented genocides, to criminal convictions invented against real, named, living people.
Constitution Failures:
🌎 tinyurl.com/mt62p9pv
The document itself:
📃 tinyurl.com/56w92b9u
















