Teachers Embrace AI, Wispr Flow Raises $30M to Kill the Keyboard

A new Gallup and Walton Family Foundation poll finds 60% of U.S. K-12 public school teachers used AI tools during the last school year. Educators are turning to ChatGPT and similar platforms to generate quizzes, worksheets, lesson plans and assist with grading.

The survey of more than 2,000 teachers reports that weekly AI users save about six hours per week and 80% credit the technology with trimming administrative chores. Roughly 60% also say it improves the quality of customized student materials and feedback.

Rapid uptake is forcing districts to craft rules while half of teachers worry student reliance on AI will weaken critical thinking and persistence. Twenty-four states now publish AI guidelines, but researchers note adoption and oversight vary widely between schools.

Dictation app Wispr Flow raised $30 million in Series A funding led by Menlo Ventures, with NEA, 8VC and several tech founders participating. Menlo partner Matt Kraning joins the board, bringing the startup’s total financing to $56 million.

The company launched a Mac app in October 2024, a Windows version in March 2025, and an iOS app this month, covering 104 languages. Monthly users are growing 50% and are split 40% in the U.S., 30% in Europe, and 30% elsewhere.

VCs were already heavy users, generating inbound interest that convinced CEO Tanay Kothari to raise capital despite approaching profitability. The funds will scale the 18-person team, launch an Android app, and add enterprise features before larger tech rivals move in.

New campus surveys reveal a widening gap between student demand for responsible AI instruction and professors’ readiness to provide it. Despite widespread student use of ChatGPT and similar tools, most instructors lack clear policies and feel ill-equipped to guide AI-assisted assignments.

At UC Davis, 70% of undergraduates want more AI training while only 33% of instructors have incorporated the technology into courses. A national survey shows just 14% of postsecondary instructors feel confident using generative AI in the classroom, leaving guidance sparse and inconsistent.

A large Turkish study cited in the article found students using unrestricted ChatGPT performed 17% worse than controls when later tested without AI, highlighting how unsupervised reliance can erode learning. Because AI use is hard to detect, the author argues bans merely push misuse underground and underscores the need for clear, assignment-level guidelines.

Only 10% of the workforce is proficient in AI – and their companies had a big role in making that happen. Section is bringing together top AI-focused leaders to help make sure your team doesn't fall in the 90%.

On July 17 12-4pm ET, join AI changemakers, business leaders, and enthusiasts to learn how to build an AI-native team.

Hear from heads of AI at some of the world’s leading companies on what went right in their AI deployments and how you can replicate that success in your own team at Section’s AI Strategy Summit.

Registration required. No charge to attend.

The article argues that physical AI—systems that interpret sensor data from the real world—should augment small human teams rather than replace them with robots. It links this shift to severe labour shortages and the rising complexity of industrial systems.

Trillions of sensors already blanket factories, vehicles and infrastructure, yet most of their data goes unused because too few workers can synthesise it in time. The author points to energy, manufacturing and construction scenarios where AI fuses weather, pressure, camera and safety readings to deliver real-time predictive alerts.

Unused sensor streams leave operators blind to impending failures and safety threats, the piece warns. AI-driven synthesis lets one expert oversee equipment that once required a full crew, directly easing the skilled-labour crunch.

University of Amsterdam scientists have mapped unique brain activations that automatically judge whether a scene invites walking, swimming, climbing, or stopping. Their findings reveal a biological blueprint for “affordances” that current AI systems consistently miss.

Thirty participants viewed indoor and outdoor photos in an MRI scanner and labeled each scene with possible movements. GPT-4 and other vision models struggled to predict these human choices, and their internal calculations did not align with the measured brain patterns.

The research confirms that affordances are encoded automatically in the visual cortex while remaining absent from leading AI models. This mismatch, the authors say, limits AI performance in physical tasks like robotics and self-driving navigation.