News You Can Use

Edition 12 · 15th - 28th Feb 2025

News You Can Use

Deep Dives

Three stories worth sitting with

Vals AI Benchmarking Report

Vals AI Benchmarking Report

What
The Vals Report evaluates the performance of 4 legal AI tools - CoCounsel (Thomson Reuters), Vincent AI (vLex), Harvey Assistant (Harvey), and Oliver (Vecflow) across 7 common legal tasks. These tasks include Data Extraction, Document Q&A, Document Summarisation, Redlining, Transcript Analysis, Chronology Generation, and EDGAR Research. The study benchmarks each tool's performance against a control group of lawyers (the Lawyer Baseline). Key findings: - Harvey Assistant: Participated in six tasks, achieving top scores in five and second place in one. Outperforming the Lawyer Baseline in four tasks. - CoCounsel: Evaluated in four tasks, it consistently ranked among the top performers, with scores ranging from 73.2% to 89.6%. Highest score for Document Summarisation (77.2%). - Lawyer Baseline: Outperformed AI tools in two tasks and matched in one. In four tasks, AI surpassed the Lawyer Baseline. Task-specific insights: - Document Q&A: Harvey Assistant achieved a high score of 94.8%. - Chronology Generation: Both Harvey Assistant and the Lawyer Baseline scored 80.2%. The study concludes that legal AI tools provide significant value in legal work, particularly in tasks involving document analysis, information retrieval, and data extraction. Certain areas such as complex research tasks still require further development to meet law firm expectations.
So what
This study provides a useful benchmark for evaluating our own AI tools such as Deal Clarity. Some key areas relevant to our workflows are document analysis, Q&A and extraction that we are actively trying to deal with. Our clients will be interested in this report, where the focus on refining human-AI collaboration is key in assessing AI strengths and gaps, and may help select suitable vendors.

Lexis Nexis Report

Lexis Nexis Report

What
LexisNexis has released a report on how legal service providers can improve client experience through technology and innovation. It highlights key client demands such as: Greater transparency in legal services; Faster turnaround time; Personalised legal service delivery. This Report emphasises that law firms and legal departments that fail to adopt AI and automation risk falling behind as clients increasingly expect tech-enabled efficiency.
So what
This Report reinforces the importance of integrating AI into client-facing workflows, not just internal use. The demand for this aligns with our work, such as exploring client dashboards that provide real-time updates (i.e. HighQ, PowerBI), refining our contract review, extraction, and analysis (i.e. Legora, Kira).

Legal Transformation Drivers - Mark Cohen

Legal Transformation Drivers - Mark Cohen

What
The Forbes article examines the ongoing transformation of the legal industry, driven by: AI adoption with automation reshaping contract review, compliance, and strategy; The rise of alternative legal service providers (ALSPs) which are using AI to offer cost-effective solutions that challenge traditional firms; Regulatory challenges as AI in legal services raises ethical and compliance concerns. The article underscores that firms must embrace AI and alternative service models or risk being left behind.
So what
The growing role of AI in legal services reinforces the need to continuously refine and expand our tools. For example, our work in developing and optimising AGPT becomes even more critical, especially with evaluating whether our AI models are delivering the depth of insight and efficiency that clients and lawyers expect. Another focus is assessing how our tools compare to AI-powered ALSPs and competitors, spotting any potential gaps in customisation, engagement and integration with existing workflows.

Worth Reading

Everything else worth a click

Vals AI - VLAIR Benchmark

The first proper independent head-to-head of CoCounsel, Vincent AI, Harvey Assistant and Vecflow's Oliver across seven legal tasks. Harvey topped five of six, and AI beat the lawyer baseline in four tasks.

Mike McGinn - Efficiency Isn't Enough

Post-R1 strategy piece warning that efficiency gains trigger more demand (Jevons' paradox). McGinn's prescription: right-size models, resist chasing the newest release, lean on batch and retrieval optimisation.

Microsoft - Majorana 1 Quantum Chip

Microsoft unveils the first quantum chip built on a Topological Core architecture using Majorana particles. Pitches a path to 1M qubits on a single chip and useful quantum machines "in years, not decades."

LawNext - AI Smackdown: Law Librarians Weigh In

Law librarians test Lexis+AI, Westlaw Precision AI and Vincent AI on the same research tasks. Distinct strengths across the three, but all work best as starting points rather than final answers.

Grok 3 Debuts

xAI released Grok 3 with extended reasoning and deep-search capabilities alongside a big marketing push. Capability gap with the frontier is narrowing, but data, governance and safety concerns still slow most enterprise adoption.

Legal Ops 101 - The Low-Down on Legal Bots

Practical primer on deploying legal bots in Slack and Teams for knowledge queries, vendor contract oversight and document organisation. Good starter if you're building your first in-house legal ops automations.

Ed Zitron - The Long Con

Zitron argues GenAI is a financially unsustainable bubble built on hype, burning billions with no path to profitability. Strong contrarian read whether or not you agree with the conclusion.

Tyler Cowen - Why AI Take-Off Is Relatively Slow

Cowen argues unproductive, change-resistant sectors dominate the economy and humans will become the bottleneck as AI gets better. Historical diffusion patterns support a slower ramp than the AGI hype suggests.

Anthropic - Claude 3.7 Sonnet with Extended Thinking

Anthropic's first hybrid reasoning model - can answer normally or switch to extended thinking mode for harder tasks. Closes much of the gap with o1/o3 on complex reasoning while staying strong on code and writing.