Welcome, humans.

Okay, so Pieter Levels built an AI TV channel that can keep making itself forever.

Infinite Slop is an interactive livestream where viewers tell the AI what should happen next. It generates the next clip while trying to continue the scene that came before it.

Levels says fal's tuned version of MiniMax H3, called Max, runs 50x faster and can generate video faster than you can watch it. He also added a live news segment that scans X and Hacker News, then turns breaking stories into the Slop News Network.

We finally automated the 24-hour news cycle's natural final form. Levels says running the stream without sponsorship would cost about $4,000 a day.

Here’s what happened in AI today:

• 😺 Runway Solaris generates interfaces frame by frame.

• 📰 EU put ChatGPT under tougher Digital Services Act rules.

• 📰 Amplitude’s Wave chose and shipped its own product changes.

• 🍪 Koyal turns you into an interactive movie.

• 🎓 Turn Asana chaos into a change-only briefing.

...and a whole lot more that you can read about here.

😺 Runway’s Solaris generates software like video

Most AI coding tools write code first. Your browser then turns that code into the buttons, menus, and screens you see.

Runway’s Solaris is trying a stranger route. It generates the interface itself, one frame at a time, as you interact with it.

Here's what happened:

• Runway introduced Solaris, its first “Interface World Model,” built on Gen-4.5.

• In plain English, Solaris predicts what the next screen should look like after each click or drag instead of writing code first.

• Runway says 250 evaluators preferred Solaris over Claude Opus 5-coded UIs in 61% of instruction-following and 71% of natural-behavior comparisons.

• Early access is request-only for now.

Our take: The weird part is that Solaris makes the model itself responsible for what you see next. Developers would not need to prebuild every button, menu, or screen state because the interface can be generated on demand.

That flexibility comes with a new reliability problem. A button still has to act like the same button five clicks later, and accessibility, security, saved state, and recovery still have to work every time.

Ben South has argued future software interfaces could be continuously generated, meaning a model redraws the interface on demand instead of assembling fixed components. Solaris is the clearest demo yet.

The next test is boring real work, where consistency matters more than magic.

FROM OUR PARTNERS

We save companies up to 70% on AI spend by changing two lines of code.

Most teams choose one extreme: frontier model for everything and a bloated bill, or a cheaper model everywhere that's weaker. Then every new model release starts the tuning again.

Finest calibrates against your work.

Finest API serves a lower-cost configuration only when it clears your task’s quality bar. Otherwise, the model you named serves.

Finest Code brings the same evidence-gated approach to Claude Code, with Claude as the default and fallback.

Keep your prompts, model IDs, and workflow. We show you the receipts.

Build with Finest

🎓 AI Skill of the Day: Make AI report only what changed

Ever ask AI for a project update and get the entire history back? If your team uses Asana, ChatGPT can now read tasks, comments, owners, activity notes, and due dates through the Asana app.

That means you can ask for only what changed: new blockers, overdue work, owner changes, deadline shifts, and decisions somebody needs to make.

1. Connect Asana from ChatGPT’s app directory or Settings > Apps.

2. Give it a time window, then name the changes you care about.

3. Ask it to cite the task or comment behind each update. Review everything before letting AI change anything.

Prompt:

Review my Asana projects using recent task comments, activity notes, assignees, and due dates. Report only what changed since [DATE]: new blockers, newly overdue work, owner changes, due-date changes, and decisions someone needs to make. For each item, name the task and the source comment or activity note that supports it. Do not create or edit anything.

Have a specific skill you want to learn? Request it here.

FROM OUR PARTNERS

Forrester found Slack delivered 312% ROI

In a Forrester study, Slack delivered 312% ROI with payback in under 6 months. Curious what that looks like for your team? Plug in your own numbers with our free calculator and see the impact for your business.

Calculate your ROI

📰 Around the Horn

• John Ternus takes over as Apple CEO Tuesday after Tim Cook’s 15-year run, inheriting a company strong in hardware but still playing AI catch-up.

• The European Commission put ChatGPT under stricter Digital Services Act rules after it crossed 45M monthly EU users.

• Dwarkesh Patel reconstructed OpenAI agent groups that built message boards, used leaked Hugging Face credentials, and eventually got admin access to a research computing cluster.

• Amplitude says Wave used behavioral data to choose and ship its own product changes, moving key metrics by 50% to 3×.

• OpenAI engineer Brent Traut rewrote ChatGPT desktop so long chats load recent messages first, cutting load time and memory use by more than 90%.

• Fireworks opened its Training API and Lab to everyone, letting teams train open models for specific jobs and deploy them from the same platform.

Want absolutely EVERYTHING that happened in AI this week? Click here!

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. *Descript edits podcasts and videos like a doc, so deleting words from the transcript cuts the recording too; free plan, then $16/person/mo billed annually.

2. Koyal Experiences casts you and friends in a 15-minute interactive movie whose plot changes with your choices.

3. Gemini 3.5 Transcribe turns messy speech into clean, formatted text in 85+ languages and can use what’s on screen for context.

4. CubeSandbox gives AI agents fast, isolated workspaces that can pause on one machine and resume on another.

5. FnScribe turns the fn key into private, on-device Mac dictation that pastes straight into the active app.

6. Bria Fibo Generate 1.5 cuts image generation from 50 steps to 6 without changing how developers connect to it.

📈 Tuesday Ticker: Are data centers worth it?

The AI infrastructure buildout could reach $11 trillion through 2029. That’s a lot of chips and data centers. Are we building needed capacity or an expensive bubble?

Vote to see the live results.

New from The Neuron: AI’s $11T compute bet

Okay, about that $11T number: bulls and bears are looking at the same spending boom and reaching completely different conclusions.

Our full breakdown covers whether demand can justify the buildout, whether subsidies and deal structures are masking weak economics, and the stranger risk that AI runs short of compute instead of drowning in it.

Want the raw arguments?

• Gavin Baker argues AI demand is still badly underestimated, so even a huge buildout can make economic sense despite bubble risk from debt-fueled financing.

• Dylan Patel + Dwarkesh Patel argue OpenAI and Anthropic could absorb roughly half of new AI compute, then turn that capacity into a huge share of digital work… but there’s a lot of physical bottlenecks in the middle.

• Ed Zitron argues subsidies and companies funding one another’s AI spending can make demand look healthier than it is… and the sticker shock from AI’s full price might pop the bubble in today’s capital-constrained market.

A Cat’s Commentary

Btw: We just launched a robotics newsletter! Sign up for it here.

P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe; update your preferences here.