Teaching legal AI @deepjudgeAI • lecturer @newenglandlaw • attorney in MA • cipp/us • former @thomsonreuters @casetext @LEAP4Lawyers #SmallLaw • my thoughts
Someone: So, what really do you do all day?
Me: Well, I read. I read carefully, and then I write carefully. And if I'm careful enough, I help someone. When I can't help, it's hard. When I do, it feels really good. That's about it. That's law work. #lawwork
@ArentFoxSchiff is the latest AmLaw 100 firm to choose DeepJudge for their AI and knowledge strategy. 650+ lawyers advising on complex, high-stakes matters are turning the firm’s deep institutional knowledge into a strategic asset.
AI learning from Star Trek…
Context rot happens when LLMs get too much info leading to errors, hallucinations
Rewatching Voyager s6e9, where Seven started concocting a range of conspiracies, the Dr's diagnosis sounded familiar...
The show was always way ahead of its time🖖
If your firm’s AI doesn’t understand context, it doesn’t understand your work.
DeepJudge connects what your lawyers know - clauses, names, concepts, meaning - into one unified view.
Know what your firm knows → https://t.co/4pwpPbtmt1
Time flies when you're having fun... 🇺🇸/🇬🇧edition!
April 2025? Teams call with a UK colleague about good non-legal uses for CoCounsel
April 2024? Staying in Vauxhall near Swift's Black Dog and introducing CoCounsel with UK colleagues
⏱️also a long stroll around London!
Waves of legal AI
🌊 1️⃣ Optimization
Enhance existing workflows by making them faster and more efficient
🌊 2️⃣ Re-engineering
Rethink processes, staffing models, pricing
🌊 3️⃣ New Business Models
Envision tech services, biz lines, reshaped value chains
https://t.co/ZfEc5yZb7Q
“@thomsonreuters tested GPT‑4.1 with CoCounsel… they found the model to be highly reliable at maintaining context across sources and accurately identifying nuanced relationships between documents, such as conflicting clauses or additional supplementary context…”
https://t.co/7PQnrwXUiP
“A new series of GPT models featuring major improvements on coding, instruction following, and long context—plus our first-ever nano model.”
"If you work in say legal, finance, engineering, insurance, this one's for you. For example, [@OpenAI] recently partnered with @thomsonreuters to use reinforcement fine-tuning to fine-tune 01 mini..." https://t.co/0Jn4znHxut | More about CoCounsel https://t.co/QYQZ1ckDzo
Which CoCounsel skills have you found helpful for innocence work? The summarize and timeline tools are very helpful, but also compare documents, which allows us to understand differences and consistency across testimony. @MSemanchik@TheInnCenter#SYNERGY2024
Exoneration is hard. For example, we were able to get 43 convictions reversed from 1999-present. The average time served under a wrongful conviction is 16 years, and we use tech to see how we can get that down to 10, 5, or less. @MSemanchik@TheInnCenter#SYNERGY2024
Before, we had to convert the legal research questions we wanted to answer into keyword guesses or Boolean queries, but now, with AI, we can start with the question. Rachel Beithon, Senior Product Manager at @thomsonreuters#SYNERGY2024
Don’t assume AI will assume—meaning, consider what your prompt is asking or instructing, what files you have given to the AI, what content is in those files, and what content is not in those files. Ken Lam, AI & prompting engineer at @thomsonreuters#SYNERGY2024
Agentic workflows—powered by always-improving LLMs—can save up to 50% of the time needed to complete common due diligence tasks. @dfwong, CPO at @thomsonreuters#SYNERGY2024
CoCounsel 2.0 offers a more responsive GenAI assistant, integrations for using your information, improved capabilities—its “skills”—and increased speed, up to 3Xs faster. @dfwong, CPO at @thomsonreuters#SYNERGY2024
CoCounsel 2.0 responds to 3 priorities identified from conversations with our users:
• Faster
• Smarter
• More capable
@dfwong, CPO at @thomsonreuters#SYNERGY2024