The Rundown, 333 Seymour Street, Vancouver, BC (2025)

11/28/2025

MIT just quantified the AI job scare: 11.7% of the U.S. labor market is now economically replaceable by existing AI, representing roughly $1.2T in annual wages across finance, healthcare, admin work, and white-collar services.

The estimate comes from the new “Iceberg Index,” a large-scale simulation that models 151M U.S. workers, mapping 32K skills across 923 occupations.

Researchers contrast a "Surface Index" of current AI use, 2.2% of wages, mostly in tech, with an 11.7% "Iceberg Index" of tasks already automatable.

Not surprisingly, the jobs most exposed are heavy on repeatable cognitive and admin work in sectors like finance, healthcare, and professional services.

The 11.7% figure represents what AI can handle right now, not what will actually get automated, that depends on corporate strategy, political decisions, and adoption speed. But some states are already using the index to run what-if scenarios on retraining and investment before the automation wave hits for real.

Source: The Rundown AI Newsletter

11/28/2025

Chinese parents are embracing a new trend: using AI assistants as live homework monitors.

Apps like ByteDance’s Doubao offer a video-call mode where parents simply point the camera at their child and tell the AI to watch for distractions.

Doubao then issues real-time prompts such as “Stop playing with your pen,” “Sit up straight,” “No sleeping,” and “Focus on your homework.”

The tool uses posture detection and behavior recognition to keep kids on task, and competing Chinese AI apps now offer similar supervision features.

While some parents love the automated discipline, others say it risks turning study time into constant surveillance, because nothing says childhood like being nagged by both your parents and an algorithm.

Source: Restofworld.org

11/28/2025

This guy built an AI device that orders food every time your stomach growls

Inventor Sohan M Rai calls the belt-clip device MOM (Meal Ordering Module), which uses improvised hardware, including a stethoscope from his sister, paired with Claude AI to detect hunger from stomach sounds and automatically place a order on a food delivery app called Zomato.

Source: Mint

11/28/2025

Shot at Beijing’s Olympic Forest Park, the 20-second clip shows the $16,000 robot running, falling, and popping back up almost instantly, movements that look surprisingly quick for a machine.

The G1 uses 43 joints, powerful lightweight motors, and and AI-driven balance from 3D LiDAR and IMUs—likely trained via Nvidia’s Isaac Simulator.

Video: UnitreeRobotics

11/28/2025

OpenAI just revealed that its analytics vendor Mixpanel suffered a security incident, with an attacker exporting some of its API users’ profile information — although no chat data, API keys, payment details, or credentials were compromised.

The breach occurred on November 9, covering Mixpanel’s systems that provided web analytics on the frontend interface of OpenAI’s API product.

The data the attacker exported included profile information associated with the API product, such as names, emails, locations (city/state), and device details.

OpenAI confirmed that users of ChatGPT and other products were not impacted, and no chat, API data, credentials, or payment details were leaked.

It removed Mixpanel and is notifying affected users directly, while urging vigilance against potential phishing attempts that could use the leaked data.

While OpenAI’s defenses held, this incident serves as a stark reminder of the security risks third-party partners can introduce. For affected API users, the immediate danger isn’t account compromise but rather social engineering, where attackers may use the leaked emails to create even more trouble.

Source: The Rundown AI Newsletter

11/28/2025

Disney CEO Bob Iger said during the company’s latest earnings call that Disney+ will soon allow subscribers to create and share AI-generated content, marking the platform’s biggest overhaul since its 2019 launch.

The upcoming features will let users generate short-form videos using licensed Disney characters such as Darth Vader, Iron Man, Anna and Elsa.

The shift comes as unlicensed AI videos of major franchises surge across platforms.

Source: IGN
Image: The Standard

11/28/2025

DeepSeek just released DeepSeek-Math-V2, an open-source MoE model that achieves gold-medal performance at IMO 2025, democratizing “research-level” mathematical reasoning that was previously locked behind proprietary walls.

The model scored 118/120 on the 2024 Putnam competition (beating the top human score) and solved 5 of 6 IMO 2025 problems, hitting the gold standard.

On IMO ProofBench, it hit 61.9%, nearly matching Google’s specialized Gemini Deep Think that won IMO gold and crushing GPT-5, which scored only 20%.

Math-V2 uses a generator-verifier system where one model proposes a proof and another critiques it — instead of rewarding final answers only.

The verifier assigns confidence scores to steps, forcing the generator to refine weak logic and ensuring step-by-step self-debugging of reasoning.

Source: The Rundown AI Newsletter

11/26/2025

GeoSpy, a visual-geolocation AI, is drawing attention for its ability to locate where a photo was taken using only the pixels inside the image, even tiny reflections.

The system ignores metadata entirely and instead analyzes colors, architecture, vegetation, shadows and environmental cues to estimate coordinates.

In demos shared online, GeoSpy was able to identify locations from just a few pixels captured in mirrors, windows or glossy surfaces, matching them against massive visual datasets.

Source: CEO Daniel Heinen

11/26/2025

Anthropic published new research analyzing 100K Claude conversations to track AI’s productivity gains, estimating that widespread AI adoption could boost annual U.S. labor productivity growth by 1.8% — doubling the current rate.

Anthropic researchers fed 100K anonymized conversations through its Clio privacy tool, mapping tasks to federal labor data to calculate productivity gains.

Researchers found Claude cuts task completion time by roughly 80%, with the average work request taking about 90 minutes without assistance.

Software developers account for 19% of estimated productivity gains, followed by operations managers, marketing specialists, and customer service roles.

Examples of tasks with massive time savings included curriculum development (96%), research assistance (91%), and executive admin functions (87%).

There is plenty of debate over AI’s actual impact vs. hype, and this research shows the real gains across a variety of sectors and tasks. But the bigger question the study sidesteps: whether the estimated doubling of productivity growth comes with the job displacement Anthropic's own CEO continues to warn about.

Source: The Rundown AI Newsletter

11/26/2025

Google’s Gemini 3 Pro has taken the top position on TrackingAI’s offline IQ benchmark, scoring 130—the highest average across the last seven tests.

The results chart shows performance distributions for 17 verbal and 10 vision models, including OpenAI’s GPT-5.1 lineup, Anthropic’s Claude 4.5 models, xAI’s Grok-4, Meta’s Llama 4 Maverick, DeepSeek V3, and Mistral Medium 3.1.

Gemini 3 Pro’s verbal and vision versions both rank near the top,surpassing Grok 4 Expert Mode, which holds a score of 126.

TrackingAI’s offline test stands out because its puzzles have never been on the public internet, giving a clearer sense of how well these models can truly reason.

Source: TrackingAI.org

11/26/2025

Startup Fleet Space says it has uncovered a far bigger lithium deposit in Quebec by combining satellites and AI, a breakthrough that could drastically speed up how we find critical minerals.

Normally, companies spend years drilling to confirm a discovery. Fleet Space instead uses a small network of satellites that scan the ground using electromagnetic and gravity sensors, then feeds that data into AI software that suggests new drill sites in just 48 hours.

The tech is now being used on the Cisco lithium project, which is estimating up to 329 million metric tons of lithium oxide. Fleet Space says the deposit could be even larger than current maps show.

Source: TechCrunch
Image: Fleet Space Technologies

11/26/2025

A $1.6 million Deloitte healthcare report for the Canadian province of Newfoundland and Labrador is under scrutiny after an investigation by The Independent found evidence of potentially AI-generated errors.

The 526-page report included false citations, references to non-existent academic papers, and misattributed research involving authors who say they never worked together.

Deloitte says AI was not used to write the report but was “selectively used” to support some citations and claims the errors don’t affect conclusions.

This marks the second controversy this year, following a similar Deloitte report for the Australian government that also contained AI-related hallucinations.

Source: Fortune
Image: Shutterstock

The Rundown

11/28/2025

11/28/2025

11/28/2025

11/28/2025

11/28/2025

11/28/2025

11/28/2025

11/26/2025

11/26/2025

11/26/2025

11/26/2025

11/26/2025

Address

Website

Alerts

Contact The Business

Shortcuts

Share

Category