Odyssey levels up world models with a playable video game

Welcome back, Superhuman. Odyssey just shattered the ceiling of what world models can do, dropping two models that generate sound and multiplayer abilities on top of real-time visuals. The lab also took a unique approach to its demo: a playable 90s-style Nintendo game (shooting game, viewer discretion advised).

Today: Odyssey ships two world models, how to use iMessage as an AI assistant for all your apps, and get the latest prompts and trending social posts.

1. Odyssey drops two world models in one day: The AI lab’s release pushes world models into audio and multiplayer territory. Agora-1 lets multiple human or AI participants interact inside the same simulation in real time — try it via a playable GoldenEye-style game. The second model, Starchild-1, generates audio alongside its visuals, and is being billed as the first real-time multimodal world model. See how it works here.

2. Apple bets on auto-deleting chats to differentiate Siri: Apple's revamped standalone Siri app (debuting in iOS 27) will reportedly let users auto-delete chat histories after a set time period. This privacy-first approach would set Apple apart from its rivals, which typically retain conversations unless you specifically opt out. Expect the new Siri to drop at WWDC 2026 on June 8.

3. Cursor ships Composer 2.5 and teases a SpaceXAI collab: Cursor just dropped Composer 2.5, its most powerful in-house coding model yet, matching frontier models on key benchmarks at a fraction of the price. It’s live now with doubled usage for the first week. The AI leader also announced that it’s working on a “significantly larger model” with SpaceXAI using 10x more compute.

The generative media landscape is shifting. In the 2026 'Future of AI' report, 8 industry leaders reveal why winning startups are pivoting to end-to-end agent workflows.

Learn how to leverage post-keyboard interfaces and make your authentic human taste your most defensible asset.

Download the report

Decreasing optimism. Gen Z's excitement over AI has slipped from 36% to 22% year-over-year, fueled by AI’s impact on jobs, the environment, and creativity. And that shift isn't just showing up in polls — it's playing out in tense moments at college graduations across the country.

Commencement speeches. Tavistock Group VP Gloria Caulfield called AI "the next industrial revolution" during a speech at the University of Central Florida. The students’ response? Immediate boos. Ex-Google CEO Eric Schmidt got a similar reaction every time he mentioned AI at the University of Arizona, although students had also protested his appearance beforehand over non-AI-related issues.

Why the shift? Schmidt actually acknowledged the elephant in the room during his speech: "There is a fear that the machines are coming, the jobs are evaporating, the climate is breaking, politics are fractured, and that you are inheriting a mess that you did not create." Many graduates blame AI, at least partly, for those fears.

Threading the needle. Nvidia CEO Jensen Huang had a different experience, with no boos or pushback during his speech at Carnegie Mellon. Huang’s framing? AI isn’t coming for students’ jobs. It’s unlocking unprecedented opportunity by letting them build any technology they want.

1. Link iMessage: Text (360) 228-5583, Sauna replies with a link to connect your phone.

2. Connect your tools in iMessage: ask Sauna to connect Slack, HubSpot, Linear, or PostHog. You'll need to authorize this in the chat.

3. Create a scheduled task through iMessage: tell Sauna the cadence and prompt.

Sample prompt: "Every afternoon, scan Linear P0s, Slack, and PostHog errors. iMessage me one short digest with anything new or needing a decision. Stay silent if nothing happens so I can keep sunbathing."

For anything in the iMessage digest needing a deeper dive, tag @Sauna in Slack for context and recommended action.

Set up your Sauna and start using iMessage here

🤖 Hermes Agent: Learn how to become an agent operator with Hermes agent (similar to OpenClaw) in this viral article that’s gotten nearly 1M views.

🤦 AI Gone Wrong: Nike is getting called out for letting a blatant ChatGPT-ism get posted to its corporate X account with 9.5M followers.

🎨 Claude Design: Anthropic just doubled token usage for Claude Design across every plan, likely as a counter to OpenAI’s generous limits.

😅 Which Path? This surprisingly simple meme perfectly captures the potential lose-lose of AI’s future and explains why so many people feel anxious right now.

🤔 Model Discussion: A screenshot of AI failing to identify Pokémon racked up 2M views. But a follow-up post pushes back, arguing that the post actually highlights the growing AI knowledge gap.

📝 Best Practices: Anthropic recently posted Best Practices for Using Claude Code in Large Codebases, featuring insights from their clients.

In Braintrust's evals course, you'll build a customer support chatbot that closes the development loop, connecting evals and observability for faster iteration. This is the same workflow teams at Ramp, Notion, and OpenAI use to ship AI with confidence—no fluff, no theory. 

Register for the course at no cost.

^{* indicates a promoted tool, if any}

💵 ChatGPT can now connect to your financial accounts and dish out tailored money advice based on your income and spending habits.

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team.