Freshfields partners with King's College London on a fully-funded LLM in Law and Technology for incoming trainees, with a £20k stipend. Sets a clear bar for the tech fluency firms will expect from juniors.
Alberto Romero's sharp critique. AGI rhetoric clashes with staff churn and engagement-optimised products, and the business models don't look sustainable.
Wordsmith's new community and content hub for the emerging "legal AI engineer" role. Useful if you're building an in-house team on this boundary.
Karpathy's talk on how LLMs rewire the way software gets built - natural language as the new programming interface and agents as the new deployment unit. Essential watch for anyone shaping tech strategy.
Provocative pricing argument: stop billing the AI-automated work, keep the hourly rate for complex partner work. Frees the firm to use AI aggressively without cannibalising revenue.
AI vendor Eudia acquires ALSP Johnson Hana (300+ legal staff). First time an AI company has bought a people-based ALSP, creating a blended AI-plus-human model for in-house legal.
Grok 4 launch and benchmark
xAI launched Grok 4 with reasoning, tool use and a larger context window. Claims top benchmark scores on several reasoning tests, though data governance and safety posture continue to hold most enterprise legal teams back.
ETH Zurich and EPFL release a fully open 8B and 70B LLM trained on the Alps supercomputer across 1,000+ languages. Apache 2.0 licence, open weights and open data - strong European option.
Case study of Legora running on Claude. Useful datapoint on how legal AI vendors pick and switch between underlying foundation models.
CoCounsel Knowledge Search pulls across HighQ, iManage, NetDocuments, SharePoint and Westlaw in a single legal workflow. Big competitor move against the "one search across everything" pitch from DeepJudge, Legora and others.
New accelerator offering up to $250k plus mentorship for early-stage legal tech founders. Another signal that capital into the category is not slowing.
BI's coverage of the Eudia/Johnson Hana deal and General Catalyst's thesis behind funding it. Puts the AI-plus-human legal services model on the investor map.
Surprising RCT finding: 16 experienced developers using AI tools took 19% longer to complete issues while believing they had gone 20% faster. Big wake-up call on subjective vs measured productivity.
AI at Work 2025 - BCG (PDF)
[Internal AG resource] BCG's 10k+ employee survey on GenAI adoption. Only 51% of frontline employees use AI regularly, 36% feel adequately trained, and "shadow AI" is widespread.
Twelve-point checklist covering culture, talent-business alignment, non-traditional recruitment, digital dexterity and AI collaboration. Useful structure for HR conversations on the AI talent model.
Rare joint warning from OpenAI, DeepMind and Anthropic researchers that chain-of-thought monitoring - the main way we understand what frontier models are "thinking" - may not survive the next generation. Real safety and assurance implications.