|
|
|
HIGHLIGHTS
Early career workers hit by the “AI effect.” A new Stanford and ADP analysis finds that employment for 22–25 year olds in the most AI-exposed occupations has fallen since late 2022, while employment for older workers in those same roles has grown. The declines are concentrated where AI automates tasks rather than augments them, raising near-term questions about how young professionals will build careers in a changing labor market. Working paper (PDF)
|
UBS scales GenAI enterprise wide. UBS logged more than 8 million AI prompts in Q2 2025, rolled out 55,000 Microsoft 365 Copilot licenses with plans to extend to all employees, and now has 52,000 staff using its in-house assistant “Red.” The bank reports more than 280 AI use cases in production and is formalizing adoption through an AI Center of Excellence and a network of “AI Champions.” These initiatives reflect a shift from experimentation to systematic integration of generative AI in daily work. ICTjournal (in French)
|
The productivity paradox before the payoff. New research on U.S. manufacturers shows that AI adoption typically brings a short-term dip in productivity before improvements arrive, reflecting the operational “J-curve.” The findings highlight that organizations must expect friction and invest in redesigning work to unlock long-term benefits. For executives, this paradox underlines the importance of patience and strategic alignment in AI rollouts. MIT Sloan
|
OTHER NEWS
- Why 95 percent of GenAI pilots fail to scale. Many initiatives stall in pilots without measurable ROI, underscoring the need for sharper problem selection, data readiness, and change management. Fortune
- China’s universities embrace AI fast. MIT Technology Review reports campus-wide adoption from AI-infused coursework to institutional assistants, outpacing the development of ethics and governance frameworks. MIT Technology Review
- AI designs new antibiotics. MIT researchers used generative AI to invent two potential antibiotics effective against drug-resistant gonorrhoea and MRSA. The study, published in Cell, shows how AI can design novel compounds atom by atom, opening the door to a new “golden age” of antibiotic discovery. BBC
|
|
HIGHLIGHTS
AI that improves itself. From self-training and tool-use to automated evaluation and iteration, a new wave of methods lets models critique and upgrade their own performance. These approaches could accelerate progress by reducing reliance on human fine-tuning and enabling systems to evolve continuously. The benefits are clear for productivity, but the risks are equally significant, as self-improving systems require strong oversight to avoid unexpected behavior. MIT Technology Review
|
China’s national AI Plus plan. Beijing’s AI Plus Action Plan aims to weave AI into every major sector including science, manufacturing, services, and public administration. The strategy calls for scaling compute infrastructure, building domestic talent pipelines, and developing industry specific AI platforms. It represents one of the most ambitious national AI agendas to date, positioning China as a driver of both technological and economic transformation. Article
|
GenVI: Co-evolving “eyes and brains.” MIT’s Generative Visual Intelligence, or GenVI, co-designs synthetic “eyes” and “brains” through evolutionary search, creating perception systems that are more capable than traditional models. By optimizing both sensors and algorithms together, GenVI can unlock new efficiencies and accuracy levels in vision tasks. The approach opens a path toward perception systems that adapt organically, more closely resembling how natural intelligence evolved. MIT publication
|
OTHER NEWS
- Interpretability moves to the critical path. Anthropic’s Dario Amodei argues that understanding model internals is now essential for safety, reliability, and regulatory compliance. Blog article
- Attribution Graphs for vision models. New work charts how inputs propagate through vision networks to specific outputs, promising clearer debugging and trust. Transformer Circuits
- When physics beats data. In climate prediction tasks, physics-informed models can outperform black-box deep learning, showing that domain knowledge combined with AI often wins. MIT News
|
|
HIGHLIGHTS
Podcast — Age of Intelligence, Ep. 5: Demba Ba on interpretability and control. Harvard’s Demba Ba joins INSEAD’s Theodoros Evgeniou to unpack what it really means to peer inside model “brains,” steer behavior, and set guardrails for high-stakes use. It is a practical conversation at the frontier of explanation, safety, and governance. Link
|
HR at the heart of AI transformation. NoeSysAI highlights the crucial role HR should play in supporting AI adoption, particularly through skills development, change management, and ethical guardrails. Ensuring that people and culture evolve alongside technology is essential for responsible and effective transformation. LinkedIn article
|
OTHER NEWS
|
Upcoming webinar with Karen Hao. On September 23, Karen Hao will join Theodoros Evgeniou and Tim Gordon in a webinar on the hidden labor, power, and global impact of AI, inspired by her book Empire of AI. Webinar details
|
|
Participation at the ECB. Theodoros Evgeniou will speak at the ECB Supervision Innovators Conference 2025, contributing to the European discussion on AI in banking and supervision. ECB – event page
|
|
|
|
|
|
|
