Felix Graef’s Post

In case there was ever a question: AI is not expected to make court reporters obsolete; instead, it is transforming the profession into a hybrid model where AI tools assist human professionals. The integrity of the legal record requires a level of accuracy, contextual understanding, and legal accountability that current AI systems cannot provide on their own. 

🧠 AI Hallucinations: A Real Risk, But Not All AI Is Created Equal Following up on our last post about the risks of AI-generated misinformation, let’s talk about trustworthy AI, especially in high-stakes fields like law. That’s where VitalLaw AI stands out. Unlike generic models, it sources and cites its information directly from authoritative legal content. ✅ No guesswork ✅ Transparent sourcing ✅ Confidence you can count on Legal professionals deserve AI that’s not just smart, but accountable. Let’s build trust in tech, one citation at a time. #LegalTech #AI #Copilot #VitalLaw #Innovation #ArtificialIntelligence #LegalProfessionals #TechThatMatters

An excellent video that made an important AI scaling law (parameter count) clear as day for me: https://lnkd.in/dyM4-Qgi Essentially: - The bias-variance tradeoff exists when your model's degrees of freedom are fewer than the function it's trying to represent (like in pre-ML days, when polynomials with <10 degrees of freedom were riskily used to represent messy real-world dynamics that likely had 100+)... - When they are equal (the "interpolation threshold"), you encounter maximum risk of overfit... - But once they are more (like in 2025 where 200B+ parameters likely exceeds the sum of degrees of freedom in all meaningful dynamics in all training data by a LOT), you enter a "second descent" of variance - an unlimited space where the more parameters you have, the more choices for perfectly-fitting functions you can find, so the smaller you can likely make the sum of squares, and thus the better you can fit all test data - with no risk of overfitting as was commonly thought when models never crossed the interpolation threshold!

Space might be the “final frontier”, but the universe of AI is pretty cool too. As lawyers and their clients expand the use of AI in their various businesses, one area I’ve been keeping an eye on is parties’ discovery obligations related to AI. AI isn’t going anywhere, and, given its ever expanding commercial use, it can only be expected to grow as a major source of discoverable information. As the expectation of discovering information from companies’ AI systems grows, so does the reciprocal expectation that companies will preserve that information when they anticipate litigation. However, with only limited information about what data is discoverable in the context of AI, businesses could be placed in a double bind: over preserve or under preserve. Either choice could be expensive. I don’t pretend to know what the right rule is. What I do suspect is that, in these early days, the ability to clearly and persuasively explain the potential benefits and burdens of discovery from AI sources will carry the day in these disputes. Each decision brings more order to the frontier of AI discovery. As lawyers our job has always been to persuasively and clearly tell our client’s stories. Times and technology may change, but the job doesn’t. How will we adapt to effectively argue these new and complex issues? I don’t expect AI is the “final frontier” of discovery, but it’s definitely a frontier. Maybe finality isn’t what makes frontiers so interesting.

Advanced technology was once reserved for the largest firms with the deepest pockets. But the technologies that have truly transformed the way we work are those that are easy to access. That’s why Litera’s agentic AI, Lito, is now integrated into the tools lawyers already know and use. 🤖 At no additional cost. About time to make advanced technology available to everyone in the legal field. #LegalTech #AIinLegal #AgenticAI ➡️ See what other lawyers are saying about Lito here: https://lnkd.in/dxQkvdp5

Spot market manipulation, rumor secrecy, and more — in real time. Smarsh AI Assistant adds context and clarity to every message review. See how it transforms compliance workflows 👉 bit.ly/46iFt28

Spot market manipulation, rumor secrecy, and more — in real time. Smarsh AI Assistant adds context and clarity to every message review. See how it transforms compliance workflows 👉 bit.ly/46iFt28

AI is already changing the way legal teams work, and the numbers speak for themselves: 79% of teams are using AI today. Check out Array’s newest white paper, The State of AI and Legal Technology Adoption, to see how your team stacks up and what actionable steps you can take next. 📘 Get your copy: https://lnkd.in/g_X5gr7M #AIAdoption #LegalTech #eDiscovery #LitigationSupport #TrustArray

The Rise of Agents! Definitions, Tasks, and Getting Things Done. From the front lines of building AI-backed legal workflows, tools and agents, this presentation examines the current state and emerging trajectory of agents in legal practice. Through concrete use cases and practical insights, this discussion moves from the present to our agentic future— exploring the emerging landscape of autonomous AI agents, identifying automatable tasks, exploring implementation considerations (e.g., guardrails, ground truth), and considering reliability (e.g., probabilism vs. determinism). Seeking evidence-based insights into AI adoption? This talk will give you potential implementation strategies. Join Damien Riehl, Vice President and Solutions Champion at vLex for Legal AI: New York as he explores how AI agents are beginning to manage real legal tasks, what they can (and can’t) do today, and how to deploy them responsibly. Register here: https://lnkd.in/enACR9qW #LegalTech #LegalAI #AIAdoption

Generative AI struggles with accuracy and defensibility in regulated fields. Agentic AI fixes this by delivering transparent, traceable workflows built for compliance, accountability, and real-world legal demands. Read more: https://lnkd.in/gygaDJm6 #Dataquest #prompts #agenticAI #case #features