HigherEd AI Daily: April 12, 2026 – AI Simulates Fluid Physics 60% Better, Why the Bottleneck Is Now Decisions Not Code, Global AI Research Tensions Rise

Sunday is for the stories that reward slower reading, and today's three go deeper than most weekly news cycles allow.

Dr. Ali Green

A New AI Simulates Fluid Dynamics Across 19 Physical Domains With 60 Percent Less Error Than Previous Models

Researchers at Polymathic AI released Walrus this week, a 1.3 billion-parameter model that simulates how fluids, gases, and plasmas move and interact across 19 physical domains including acoustics, astrophysics, and non-Newtonian fluids. In head-to-head benchmarks, Walrus achieved the lowest error in 18 of 19 domains for single-step predictions and reduced one-step error by an average of 63.6 percent compared to the best competing models. The model is freely available under an MIT license.

Why this matters for your teaching

For faculty in physics, engineering, atmospheric science, and computational biology, Walrus is a research tool available today. For faculty in any discipline, it is a concrete example of how AI is accelerating scientific simulation in ways that may reshape graduate research in quantitative fields over the next five years.

Read more: The Batch by Andrew Ng (April 10 Issue)

Pulled from: The Batch by Andrew Ng

Andrew Ng: The Real Bottleneck in AI-Assisted Work Is No Longer Building. It Is Deciding What to Build.

In The Batch this week, Andrew Ng argued that as AI coding agents write software faster and cheaper, the constraint shifts from technical execution to product judgment. He described this as the Product Management Bottleneck: a growing gap between what AI can build and the human capacity to decide what is worth building. He noted that more applications will be created for smaller audiences as the cost of development falls, raising the premium on good decision-making and domain expertise.

Why this matters for your teaching

This framing has direct implications for how faculty teach research design, project management, and applied AI courses. The skill that will be most valuable is not prompting or coding, but knowing which problems are worth solving and which solutions are good enough.

Read more: The Batch by Andrew Ng (April 10 Issue)

Pulled from: The Batch by Andrew Ng

NeurIPS Reversed Restrictions on Chinese Researchers After a Boycott Threat, Signaling Rising Tensions in Global AI

The Batch reported this week that NeurIPS, one of the top AI research conferences, reversed a policy restricting participation by Chinese researchers following a threatened boycott by prominent researchers worldwide. The episode reflects growing tension between geopolitical pressures on AI development and the international character of the academic AI community. The debate touched on whether research conferences should be governed by export control logic or by the norms of open science.

Why this matters for your teaching

For faculty who publish in AI conferences, supervise international graduate students, or teach technology policy, this episode is a live case study in how geopolitics is reshaping academic norms. The conversation will intensify, not diminish, over the next several years.

Read more: The Batch by Andrew Ng (April 10 Issue)

Pulled from: The Batch by Andrew Ng

Try something new today

Prompt of the Day

"I teach a course on [research methods or science policy or technology ethics] and I want to design a one-week module around the question of how AI is changing what scientific research looks like. I want students to engage with at least one recent AI research release, one policy controversy, and one practical tool. Suggest a three-day reading and activity sequence, name the specific materials you would use, and write one discussion question for each day."

Tool of the Day

Consensus

AI-powered academic search that pulls answers directly from peer-reviewed papers. Useful for quickly finding what the research actually says on any given question.

Free (limited) / Paid