0727

AI Insights & Research

Peer-reviewed Registered Report; behavioral experimentsScientific Reports·

Does reading the advice change trust in AI ethical guidance?

This Registered Report compares GPT-4 and expert advice on ethical dilemmas, with a 187-person pilot and a 642-person experiment. Exposure to advice and source disclosure influenced willingness to accept it. The outcomes concern perceived quality and preference, not moral authority or improved consequences in real decisions.

Source
Product-method announcement and vendor benchmarksGoogle DeepMind·

Question-directed video inspection instead of uniform sampling

Google describes Gemini actively searching and revisiting video segments across frames, audio and transcripts. Vendor benchmarks report efficiency and accuracy gains in tested conditions. Maximum improvements depend on the model and task; they do not establish reliable performance for every long-video or monitoring deployment.

Source
Energy analysis and outlook reportInternational Energy Agency·

AI expansion must meet the pace of electricity infrastructure

The IEA examines updated evidence on data-centre electricity demand, grid connections and supply-chain constraints. Power, finance and community acceptance all shape infrastructure expansion. Future demand estimates remain scenario-dependent; planned investment should not be confused with operational computing capacity.

Source
Peer-reviewed methods and application researchNature Machine Intelligence·

Separating reasoning and tool execution in materials research

MatBrain assigns materials reasoning and tool coordination to two specialized models, evaluating them on structure generation, property prediction and synthesis planning. The paper reports catalyst-candidate screening. This summary uses the public abstract and figure descriptions; computational candidates should not all be treated as experimentally validated or production-ready.

Source
Organizational framework and early case analysisMcKinsey & Company·

From organization charts to task networks in human–agent teams

McKinsey proposes small human–agent teams organized around business outcomes, supported by changes in business models, operations, governance, people and data. Humans retain direction, critical judgment and accountability. This is a framework informed by early deployments; its scaling claims and capability forecasts remain conditional.

Source
Product-engineering survey and practice analysisMcKinsey & Company·

Faster AI coding requires redesigning the delivery system

Drawing on 334 product and engineering respondents, McKinsey describes uneven AI productivity gains. It treats workflows, roles, verification and change management as one delivery system, with checking capacity keeping pace with generation. These survey and practice-based findings do not guarantee similar gains for every engineering team.

Source
Self-reported enterprise surveyMcKinsey / QuantumBlack·

McKinsey: Why individual AI gains have not become broad financial returns

In McKinsey’s survey of 1,719 respondents in 97 countries, 80% reported individual productivity improvements, while 37% attributed positive enterprise EBIT impact to AI. High performers more often redesigned workflows. These are self-reported associations, not proof that a particular management practice causes higher returns.

Source
Corporate education-policy announcementMicrosoft·

Classroom AI: Protect students and preserve teachers’ judgment

Microsoft’s education commitments emphasize privacy, teacher control, student thinking and transparency, alongside a school safety standard developed with a teachers’ union. This is a corporate policy and practice announcement. Its principles and customer examples do not substitute for independent evaluation of learning outcomes.

Source
Philosophical research perspectiveMicrosoft Research·

Is AI an extension of human intelligence? A philosophical account

Drawing on phenomenology, the authors interpret AI as extending structures in human language and cognition, distinguishing fluent output from reliable understanding. They frame safety as a system-design and governance challenge. This philosophical argument should not be treated as an experimentally established ceiling on AI capabilities.

Source
Draft code of conduct for public consultationMicrosoft AI·

Microsoft’s draft AI code centers meaningful human control

Microsoft’s draft Humanist AI Code calls for models to accept correction, interruption and shutdown, and remain within human-authorized scope. Released for public consultation, it states intended training and deployment norms. It is not evidence that current models fully comply or that scientific questions about AI consciousness are settled.

Source
Stay curious. See you in the next read.

Join the conversation after approval.

Comments, replies and messages are reserved for approved, signed-in collaborators.

  1. Submit your collaborator profile
  2. Wait for profile review
  3. Sign in after approval to participate
Already a member? Sign inBecome a collaboratorThis is a flow preview. Account sign-in and approval checks are not connected yet.