weekend ai reads for 2025-10-03

October 03, 2025

šŸ“° ABOVE THE FOLD: WORLD MODELS

What Are ā€˜World Modelsā€™? The Key to the Next Big AI Leap ā€” To develop knowledge beyond text and videos, AIs must have realistic virtual playgrounds where they can make mistakes and learn / Wall Street Journal, archive (8 minute read)

ā€˜World Models,ā€™ an Old Idea in AI, Mount a Comeback ā€” Youā€™re carrying around in your head a model of how the world works. Will AI systems need to do the same? / Quanta Magazine (8 minute read)

AI groups bet on world models in race for ā€˜superintelligenceā€™ ā€” Google DeepMind, Meta and Nvidia are developing systems that aim to better understand the physical world / Financial Times, archive (7 minute read)

Unprecedented volumes of Earth observation data are continually collected around the world, but high-quality labels remain scarce given the effort required to make physical measurements and observations. This has led to considerable investment in bespoke modeling efforts translating sparse labels into maps. Here we introduce AlphaEarth Foundations, an embedding field model yielding a highly general, geospatial representation that assimilates spatial, temporal, and measurement contexts across multiple sources, enabling accurate and efficient production of maps and monitoring systems from local to global scales.

  • their dataset consists of global, annual embedding field layers, covering Earthā€™s terrestrial surface at ~10 mĀ² resolution from 2017 through 2024

  • the dataset, Earth Engine Data Catalog / Google for Developers

 

šŸ“» QUOTES OF THE WEEK

You will die mid-scroll.

Alberto Romero (source)

 

Weā€™re living through very odd times when a management consulting firm is the most sensible voice in the room.

Bryce Elder (source)

 

šŸ‘„ FOR EVERYONE

There is no way to opt out, according to Meta.

Spotify is finally taking steps to address its AI slop and clone problem ā€” In addition to removing impersonating and spammy content, the music platform is developing a standard to disclose AI use in music. / The Verge (5 minute read)

ā€˜My son genuinely believed it was realā€™: Parents are letting little kids play with AI. Are they wrong? ā€” Some believe AI can spark their childā€™s imagination through personalized stories and generative images. Scientists are wary of its effect on creativity / The Guardian (18 minute read)

  • kids are idiots and believe in Santa Claus; this is what weā€™re worried about?

ā€œWorkslopā€ was the logical outcome of productivity maxxing ā€” Local throughput optimization always externalizes costs onto downstream colleagues. / The Product Picnic, Beehiiv (6 minute read)

 

šŸ“š FOUNDATIONS

ChatGPT can buy stuff for you now - forever changing online shopping ā€” Transactions between AI agents are on the rise. Hereā€™s how to try ChatGPT's new Instant Checkout feature. / Zdnet (5 minute read)

Effective context engineering for AI agents / Anthropic blog (17 minute read)

Microsoft just added AI agents to Word, Excel, and PowerPoint - how to use them ā€” Available for Microsoft 365 Copilot users, the new agents will help you create, edit, and analyze your documents, spreadsheets, and presentations. / Zdnet (7 minute read)

 

šŸš€ FOR LEADERS

There is widespread sense AI adoption is slowing down and productivity gains are hard to get. Is it true? Why? I just attended the inaugural conference of the Center for AI, Management and Organizations at Hong Kong University. A thread on insights. / lugaricano, XCancel (6 minute read)

  • thread on expectations, obstacles, skill compression, and more

Meet ā€œjob hugging,ā€ the workplace trend thatā€™s quietly draining the bottom line of many businesses.

The result is a workplace paradox: Companies invest in AI to boost productivity even as employees actively resist using it in order to protect their jobs. That means employers are essentially paying for efficiency gains theyā€™ll never actually get.

One proposal being discussed at a major investment bank is reducing the ratio of junior bankers to senior managers from the current 6-1 to 4-1. In the new regime, half of those junior bankers would be working from cities with cheaper labor, say Bengaluru, India, and Buenos Aires, Argentina, instead of being clustered in expensive New York.

 

šŸŽ“ FOR EDUCATORS

Perplexity Comet and online quizzes / Dā€™Arcy Norman, PhD (2 minute read)

I just logged into our Brightspace environment using Comet and took a quiz. With the quiz open in the browser, I asked the AI assistant ā€œwhat are the answers?ā€

And the Comet AI thing answered all of the questions correctly, in about a minute, right in the browser.

How Math Teachers Are Making Decisions About Using AI ā€” A Stanford summit explored how K-12 educators are selecting, adapting, and critiquing AI tools for effective learning. / Human-Centered Artificial Intelligence, Stanford University (12 minute read)

ā€œThe information is tailored to a specific subject area. Itā€™s vetted. There are actual people behind it who are selecting whatā€™s included,ā€ Shah said. ā€œSources are cited. Thereā€™s no ads. People arenā€™t making money off of you. Thereā€™s no data saved at all.ā€

Mindsmith ā€” Mindsmith is an authoring tool that applies generative AI to every step of the eLearning creation process. Build beautiful, instructionally sound eLearning in a fraction of the time.

UNEvolution from Ugly Duckling to Cygnet / Future Campus (5 minute read)

The team are looking at opportunities to leapfrog the universityā€™s technological debt ā€“ the challenges of grappling with ageing legacy systems for enrollment and admissions may be bridged with a layer of AI assistants, A/Prof Driver said, but the jury is still out as to how far the issues of systems that donā€™t talk to each other can be resolved.

 

šŸ“Š FOR TECHNOLOGISTS

Our job is to design experiences that guide users away from the dangerous poles of Active Distrust and Over-trust and toward that healthy, realistic middle ground of Calibrated Trust.

Vibe code straight from your website / UX Collective (12 minute read)

By vibe coding straight from your website, you keep the design process realistic to the userā€™s experience, and keep design and implementation in sync from the start.

Claudesidian: ClaudeCode + Obsidian Starter Kit / heyitsnoah, GitHub (11 minute read)

Turn your Obsidian vault into an AI-powered second brain using Claude Code.

This is a pre-configured Obsidian vault structure designed to work seamlessly with Claude Code, enabling you to:

- Use AI as a thinking partner, not just a writing assistant
- Organize knowledge using the PARA method
- Maintain version control with Git
- Access your vault from anywhere (including mobile)

 

šŸŽ‰ FOR FUN

And Every. Single. Poster was vandalised, it literally looks like some of the most beautiful art you have ever seen

OpenAI Sora Video Generator Makes Brain Rot. It Could Also Be Huge. / Parmy Olson, Bloomberg Opinion (7 minute read)

In the app, you can create what OpenAI calls a ā€œcameoā€ of yourself by uploading biometric data. When you first join the app, youā€™re immediately prompted to create your optional cameo through a quick process where you record yourself reading off some numbers, then turning your head from side to side.

Each Sora user can control who is allowed to generate videos using their cameo. You can adjust this setting between four options: ā€œonly me,ā€ ā€œpeople I approve,ā€ ā€œmutuals,ā€ and ā€œeveryone.ā€

 

šŸ§æ AI-ADJACENT

Coasean Bargaining at Scale / Cosmos Institute blog (37 minute read)

What if, in the future, everyone had their own super-smart AI agent - like a digital assistant that really knows you? This agent would understand all your preferences: when you need quiet, what you're willing to pay for, what bothers you, and what doesnā€™t.

  • basic idea in an example: if you want to have a party and your neighbors prefer silence, they would set a price the silence is worth to them and youā€™d have to pay it

  • this misses a lot; for example, someone needs to decide the starting rules ā€” is a personā€™s right to quiet more important than anotherā€™s right to not?

  • the opportunities for abuse seem infinite

  • fun thought exercise but this idea feels like science fiction that the author hasnā€™t really thought through

 

ā‹„