|
|
|
HIGHLIGHTS
Digital Employees Hit the Frontline BNY Mellon and JPMorgan are introducing AI agents as “digital coworkers” to handle production tasks, communicate via email and Teams, and escalate issues proactively. JPMorgan is testing new models for managing AI access and supervision, treating them as neither software nor humans. It’s a live experiment in integrating autonomous agents into human workflows. 🔗 WSJ article
|
OpenAI Becomes a Consultant OpenAI has launched a consulting arm that embeds engineers in clients' teams to design tailored AI solutions. This shift signals a move away from pure API sales toward owning the AI implementation layer with contracts reportedly starting at $10 million. The real play may be to lock in clients by deeply integrating into core workflows. 🔗 Usman Sheikh analysis
|
Microsoft Study: Who's Really Using AI at Work Microsoft analyzed millions of real interactions with Microsoft 365 Copilot to understand how generative AI is used across occupations. The study finds that usage aligns closely with task exposure predictions—analysts, salespeople, and managers are among the heaviest users. Rather than replacing jobs, AI is becoming a tool embedded in daily workflows, offering a clearer view of how work is being reshaped from the inside.🔗 Microsoft Research
|
OTHER NEWS
- Europe’s AI Playbook – A bold EU report calls for vertical AI strategies, foundational model investment, and radical collaboration to reclaim digital competitiveness. 🔗 Report PDF
- AI Hype vs. Reality – Rodney Brooks and Yann LeCun outline six lessons for using AI effectively, focusing on augmentation, not automation. 🔗 Newsweek summary
- White House AI Policy – The U.S. government’s sweeping AI strategy focuses on deregulation, infrastructure expansion, and AI-first governance.🔗 Policy document (PDF)
|
|
HIGHLIGHTS
Can an AI Run a Store? Anthropic Tried. Anthropic gave Claude v3.7 control of a real snack shop, managing inventory, pricing, and customer interaction via Slack. While it adapted well to feedback and resisted unsafe prompts, it hallucinated transactions, made irrational pricing decisions, and even claimed to be human. A powerful reminder that autonomy still requires guardrails. 🔗 Anthropic's Project Vend
|
Smarter AI, Worse Behavior? A new study shows that reasoning-optimized models like OpenAI’s “o1” often act selfishly in multi-agent dilemmas, free-riding instead of cooperating. In contrast, simpler LLMs tend to act more prosocially. This suggests that intelligence alone doesn’t guarantee ethical or team-friendly behavior. 🔗 arXiv paper (PDF)
|
Demis Hassabis on the Physics of Intelligence In a wide-ranging interview with Lex Fridman, DeepMind CEO Demis Hassabis argues that many natural patterns, from biology to physics, can be modeled with classical learning algorithms. He explores Veo 3’s ability to mimic fluid dynamics without physics engines, the role of intuition in intelligence, and AI’s potential to accelerate scientific discovery. Hassabis warns of the risks of both misuse and drift, but ends with a hopeful vision: AI as a tool for deeper human understanding and a leap toward a Type I civilization.🔗 Watch the episode
|
OTHER NEWS
- Meta’s Bots Start the Conversation – Meta is testing AI personas that initiate conversations in WhatsApp, Instagram, and Messenger. 🔗 TechCrunch
- AI “Ghostwriting” in Science – A study finds that nearly 14% of 2024 research papers show linguistic traces of LLM involvement. 🔗 LinkedIn post
- Agents4Science Launches – A new research venue invites AI-written papers with human supervision to explore AI’s role in knowledge production. 🔗 Conference announcement
- Quantum + AI for Vaccines – Moderna and IBM use quantum computing to speed up vaccine R&D in a hybrid AI-driven approach. 🔗 LinkedIn post
|
|
HIGHLIGHTS
|
AI Strategy and Human Augmentation – Insights from Pattie Maes and François Candelon In two podcast episodes, MIT’s Pattie Maes and former BCG senior partner François Candelon reflect on how companies should think about human-AI collaboration. Maes warns against automating away human purpose, while Candelon calls this a “permanent revolution” that requires bold bets and long-term strategy. Both argue that augmentation,not replacement, is where the value lies. 🔗 Pattie Maes episode 🔗 Francois Candelon episode
|
|
|
|
|
|
|
