10 topics covered

Listen to today's briefing
0:00--:--

Public Sentiment: Americans Express Concerns About AI Pace

What happened: A Pew Research poll shows that while half of Americans use AI chatbots, nearly two-thirds believe AI is advancing too quickly.

Key details:

  • 49 percent of Americans report using chatbots at least occasionally
  • 63 percent believe AI technology is advancing too quickly
  • ChatGPT usage has doubled since 2023, with 44 percent of Americans now using it
  • Overall AI chatbot use increased from 33 percent in 2024 to current levels

Why it matters: The gap between adoption (50%) and concern about speed (63%) suggests public anxiety about AI progress is outpacing comfort with the technology. This sentiment may influence policy and regulatory decisions even as usage grows.

Practical takeaway: If you're building AI products for consumer markets, expect growing skepticism about rapid innovation. Emphasize safety, control, and transparency—and prepare for regulatory momentum that may follow public concern about pace.

Robots Learn Dexterous Tasks Through AI Coding Agents

What happened: Researchers from Nvidia, Carnegie Mellon University, and UC Berkeley demonstrated that robots can learn complex grasping skills in the real world using AI coding agents as their training mechanism.

Key details:

  • A fleet of eight robots achieved up to 99 percent success on complex grasping tasks
  • AI coding agents generate and refine robot control code autonomously
  • Researchers use this approach to teach dexterous manipulation in physical environments
  • Collaboration between Nvidia, Carnegie Mellon University, and UC Berkeley

Why it matters: This research shows AI coding agents can accelerate robot learning in the real world by orders of magnitude compared to manual programming. The 99% success rate on complex tasks suggests AI-generated control code can achieve production-grade reliability, unlocking scalable robotics training.

Practical takeaway: If you're developing robotics systems, consider AI coding agents as a tool for generating and optimizing control logic. This approach may reduce the manual engineering required to achieve high-reliability grasping and manipulation tasks.

AI Research Methods Critique: The Goat Neural Network Experiment

What happened: A Microsoft researcher built a functioning neural network out of goats, bridges, and ice ramps in Age of Empires II to critique how AI research evaluates model capabilities.

Key details:

  • Researcher analyzed 315 AI papers and found more than half assume language models have human-like traits before experiments even begin
  • The goat-based neural network performs mathematically identical operations to a real neural network
  • The goat version demonstrates that perception of intelligence depends on interface design, not underlying capability
  • By replacing the chat interface with wandering goats, the same mathematical functions feel unintelligent

Why it matters: This critique exposes a fundamental flaw in how AI research evaluates capability—many papers conflate the appearance of intelligence (driven by chat interfaces) with actual performance. It highlights that benchmark results may reflect interface design more than true model capability, and that anthropomorphization biases research conclusions.

Practical takeaway: When evaluating AI model benchmarks, scrutinize what's actually being measured versus what perception of capability is being created by the interface. Be skeptical of claims that cite human-like behavior as evidence of capability—the goat experiment shows the two are separable.

Midjourney Enters Medical Hardware with Full-Body Scanner

What happened: Midjourney, the bootstrapped AI image generation company, unveiled its first hardware product: an ultrasound-based full-body scanner called The Midjourney Scanner, with plans to build a San Francisco spa.

Key details:

  • The Midjourney Scanner is an ultrasound device using a ring of sensors to capture full-body scans
  • CEO David Holz acknowledged the shift from "cat pictures" to medical imaging is a significant product pivot
  • The spa project marks Midjourney's expansion beyond software into physical spaces and healthcare

Why it matters: Midjourney is signaling its ambition to move beyond image generation into vertically-integrated healthcare and wellness. This represents a major strategic pivot for a bootstrapped company and shows AI labs expanding into physical-world applications with real regulatory and liability stakes.

Practical takeaway: If you're considering Midjourney for image generation, understand that the company is diversifying its business model and engineering resources. For healthcare applications, monitor how Midjourney's scanner performs in medical validation—hardware in healthcare carries FDA scrutiny and clinical requirements.

Open-Source Coding Models: GLM-5.2 Closes Gap with Frontier Models

What happened: Chinese AI lab Zhipu AI released GLM-5.2, an open-source model under the MIT license that achieves competitive performance on long-form coding tasks.

Key details:

  • GLM-5.2 maintains a stable 1-million-token context window
  • On FrontierSWE benchmark for hours-long coding tasks, GLM-5.2 trails Anthropic's Claude Opus 4.8 by just one percentage point
  • On reasoning benchmarks, the model still lags well behind closed-source rivals
  • Released under MIT license, making it freely available for commercial use

Why it matters: GLM-5.2 demonstrates that open-source models can achieve near-parity with frontier closed-source models on specialized tasks like extended coding marathons. This narrows the capability gap and increases cost pressure on commercial providers while enabling developers to run comparable models locally.

Practical takeaway: If extended-context coding tasks are your priority, evaluate GLM-5.2 as a cost-effective open alternative. For general reasoning, closed-source models still hold an edge, but the gap is shrinking rapidly.

AI-Powered Hardware: Google Smart Speaker & Snap Glasses Launch

What happened: Google announced its first new smart speaker in six years, shipping June 25, while Snap unveiled its $2,195 Specs glasses, an AI-powered wearable designed for hands-free computing.

Key details:

  • Google Home Speaker starts shipping June 25; preorders opened June 17 at $99 price point
  • Hardware design unchanged from the announcement nine months prior
  • Snap Specs glasses priced at $2,195
  • Snap CEO Evan Spiegel described the 12-year project as an attempt to "bring computing into the world" and "make it more human"
  • Snap positions Specs to help people with hands-free interaction

Why it matters: Both launches signal AI hardware is becoming a primary battleground for consumer tech companies. Google's long-delayed speaker and Snap's premium wearables show that companies are betting on specialized AI devices as alternatives to smartphone-centric computing.

Practical takeaway: If you're researching AI hardware trends, the $99 mass-market (Google) and $2,195 premium (Snap) price points show market segmentation. Google's speaker will rely heavily on cloud AI (Gemini), while Snap's position suggests local processing capabilities. Monitor adoption of both to understand which model consumers prefer.

Major AI Talent Migration: Noam Shazeer to OpenAI

What happened: Noam Shazeer, a co-author of the foundational "Attention Is All You Need" paper and former co-lead of Google's Gemini models, is joining OpenAI after a brief two-year return to Google.

Key details:

  • Shazeer returned to Google from Character.AI in 2024 as part of a $2.7 billion acquisition deal
  • He was co-lead of Google's Gemini models before departing
  • This follows Andrej Karpathy's move to Anthropic earlier in 2026

Why it matters: The continued exodus of top AI researchers from Google to competing labs signals shifts in where the most influential AI talent sees the best opportunities. This represents the second major AI hiring shake-up this year and suggests OpenAI is winning the competition for frontier AI researchers.

Practical takeaway: Watch for patterns in where top researchers move as an indicator of which labs are viewed as leading frontier AI development. Google's loss of foundational researchers may impact its competitive position in coming model generations.

OpenAI Safety Research: Predicting Model Failures Before Launch

What happened: OpenAI researchers proposed a new method for predicting how often an AI model will make mistakes after its public release, addressing gaps left by standard safety testing.

Key details:

  • The research proposes techniques to forecast model failure rates before deployment
  • Method aims to fill gaps that traditional safety testing leaves undetected
  • Research focuses on mistakes that occur in production after launch

Why it matters: Better failure prediction could improve pre-release confidence in model safety and help companies set appropriate guardrails. This shifts safety assurance from binary "pass/fail" testing to probabilistic failure-rate estimation, which is more realistic for deployed systems.

Practical takeaway: When evaluating new AI models, ask providers whether they've used predictive failure-rate analysis and what their forecast error margins are. This methodology could become a standard part of model safety evaluation.

Major Funding Round for Odyssey ML World Models

What happened: Amazon, Nvidia, and AMD are jointly investing $310 million in Odyssey ML, a startup building 3D world models, valuing the company at $1.45 billion.

Key details:

  • Amazon, Nvidia, and AMD are leading the funding round
  • CIA-linked fund IQT and Google chief scientist Jeff Dean are also backing the round
  • Odyssey ML is developing world models—AI systems that can simulate and predict 3D environments
  • Company valuation is now $1.45 billion

Why it matters: The convergence of hardware (Nvidia, AMD), cloud (Amazon), intelligence agencies, and top researchers around world models signals industry consensus that 3D world modeling is the next major frontier after language models. This bet represents enormous confidence in world models as the foundation for robotics, simulation, and autonomous systems.

Practical takeaway: If you're building robotics or autonomous systems, monitor Odyssey ML's progress on world models—this technology may become foundational infrastructure for physical AI applications. The involvement of Nvidia and hardware vendors suggests world models will drive new chip demand.

Anthropic Export Controls Deadlock

What happened: Anthropic is caught in a government export control deadlock that has kept its Fable 5 and Mythos 5 models offline, with confusion around regulatory requirements making restoration difficult.

Key details:

  • The Trump administration ordered Anthropic to cut access for all foreign nationals, including users inside the US and its own employees
  • Anthropic has spent much of this week fighting to get models back online
  • Uncertainty persists about the actual scope and interpretation of the export rules
  • The White House-Anthropic conflict involves competing interpretations of national security concerns

Why it matters: The unclear and rapidly-changing regulatory environment creates operational chaos for frontier AI companies, preventing them from serving any customers while rules are interpreted. This sets a precedent for how export controls may disrupt AI model availability unpredictably.

Practical takeaway: If you rely on Anthropic's latest models, expect extended unavailability while the company and government negotiate rule interpretation. Companies should prepare contingency plans for sudden model access restrictions based on export policy.