Jev Legal Demos
Dan Sito (ex-big law partner, now @gen_legal_inc partner) built 4 Jev demos, including an unsped-up 83(b) one.
THE JEV GUIDE · CURATED EDITION
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
Sponsors
These products pay for the directory. Want yours up here?
Dan Sito (ex-big law partner, now @gen_legal_inc partner) built 4 Jev demos, including an unsped-up 83(b) one.
A Turkish team shares Kılavuz, a new family of decision models. Its first model, Kılavuz-Hukuk (~2B parameters), reportedly beats TypeSafe AI's law-focused JEV model across all initial tests. Link
HAQQ announced it ran TypeSafe's Jev across its legal AI stack for under $0.20, with a linked blog post on legal AI hallucination detection. Link
On a subset of CLERC queries, Laya without fine-tuning scored 2.5%, below BM25 (5%) and FastLate (10%); Jev reached 30%, Luna 22.5%, and with prompt tuning Luna gets 52.5–55%.
User shares how to solve high AI bills, recommending JEV, saying it's fast, cheap, and reads everything, e.g., a law firm can review 5,000 contracts before Friday.
One card in the feed, every sixteen, on every page of it. $50 a week, cancel any week.
The post describes an anon-mode feature that uses the Jev model to redact PII from all documents, showcasing Jev's application in data compliance and privacy protection.
The author proposes using Gemini to read copyright claims and JEV to check which file existed first, so YouTube doesn't move ad revenue until timestamps align.
The post notes that legal case classifiers existed before Jev, but emphasizes Jev's intelligence, implying value in legal applications.
The author states they are preserving evidence in case a company sues Jev over false and dangerous calibration claims.
Jozef introduces LegalJev, using Jev to verify legal citations in under 9 seconds, helping catch AI-generated fake case law and misattributed citations.
The author tested the Jev model as a Kleros juror in the ClawBank dispute, finding most of the implementation is ordinary deterministic code composing 24 numbers supplied by Jev.
The author built hakaretolcer.com, which uses the Jev AI model to analyze whether tweet text could be considered an insult, helping users avoid risk before posting aggressive tweets. Link
Star the project on GitHub to support continued updates and curation.