TRD Runway's Solaris previews the no-code internet

[admin]

.bh__table, .bh__table_header, .bh__table_cell { border: 1px solid #C0C0C0; } .bh__table_cell { padding: 5px; background-color: #FFFFFF; } .bh__table_cell p { color: #2D2D2D; font-family: 'Helvetica',Arial,sans-serif !important; overflow-wrap: break-word; } .bh__table_header { padding: 5px; background-color:#F1F1F1; } .bh__table_header p { color: #2A2A2A; font-family:'Trebuchet MS','Lucida Grande',Tahoma,sans-serif !important; overflow-wrap: break-word; } Read Online | Sign Up | AdvertiseGood morning, {{ first_name | AI enthusiasts }}, and welcome to our 4,629 new readers. Runway just previewed an internet where nothing exists until you open it, with interactive sites and apps that have zero code running under the hood. Solaris instead renders everything as real-time video, creating each frame as you click — and the internet as one long hallucination might actually be loading.In today’s AI rundown:

• Runway's Solaris previews AI era of interfaces
• AI catches heart disease in two seconds
• How to connect ChatGPT to iMessage
• Anthropic tests automated AI safety research

LATEST DEVELOPMENTSRUNWAY Runway's Solaris previews the AI era of interfacesImage source: RunwayThe Rundown: Runway just introduced Solaris in early access, an “Interface World Model” that renders websites and apps as live video, with every frame drawn in real time as a user clicks and drags with no code running underneath.The details:

• The system pairs Runway's Gen-4.5 video model with an LLM that reads each click or drag, decides what should happen, and prompts the next frames.
• Demos include a shirt try-on dragged from a virtual rack onto a photo, a salad built by dropping in ingredients, and an interactive combustion demo.
• In a Runway study, testers picked Solaris over pages coded by Claude Opus 5 in 71% of matchups judging in-scene behavior and 61% on instruction following.
• The company said the build still hit issues like text legibility, long session drift, and convincing-but-wrong screens, with Solaris heading into early access for testing.

Why it matters: Is this what the future of the web looks like, or just a novelty? Whichever side you land on, what’s enabling it is the speed, cost, and quality convergence we’re starting to see across AI models. If that curve continues, the doors open for completely new ideas and interfaces not previously realistic or cost-effective. TOGETHER WITH OPTIMIZELY How Corinthia Hotels hit 100% brand approval rateThe Rundown: 56% of consumers can spot AI-generated content. 40% say it's eroding their trust in your brand. To solve this, Corinthia Hotels built a 7-agent AI ecosystem to execute personalization across 5 personas, 14 lifestyle stages, over 75 behavioural signals, and 6,000+ content variants without sacrificing brand standards.Join Optimizely’s upcoming webinar on Oct. 7th to:

• Hear how they hit a 100% brand approval rate, achieved 10x productivity, and saved 150+ hours
• Explore the 7-agent architecture achieving zero hallucination across outputs
• Watch demos of autonomous personalization and creating dynamic 1-1 experiences

Register for Oct. 7 at 10 AM ET to join live or receive the recording.AI HEALTHCARE AI catches heart disease in two secondsImage source: British Heart FoundationThe Rundown: Imperial College London researchers introduced an AI model that reads a routine ECG in under two seconds, spotting heart failure and valve disease undetectable by doctors — with trials queued up ahead of a possible national rollout.The details:

• Confirming the diseases previously meant a months-long queue for an ultrasound, even with hospitals running over 1B ECGs every year.
• The model was trained on 10.6M ECGs and tested on 65k patients, with AI flagging heart failure successfully in 81% of cases and valve disease in 90%.
• Imperial’s Prof. Fu Siong Ng said hospitals could eventually run the model over every ECG, potentially catching issues that weren’t even being screened for.
• The team is starting with a 590-patient trial at six hospitals, with a goal of incorporating the AI into routine National Health Service use within two years.

Why it matters: We hear about AI eventually curing all diseases, but the near-term wins are currently happening in better detection methods on data already being collected. Just like we saw last week with the first AI-assisted brain surgery, a second set of expert eyes is going a long way towards better diagnosis and treatment. AI TRAINING  How to connect ChatGPT to iMessageThe Rundown: In this guide, you will learn how to connect iMessage to ChatGPT Work, discovering the capabilities, limitations, and some quick ways to make it work better.Step-by-step:

• Open the ChatGPT desktop app on your Mac. Click Plugins > search "Messages" > install the plugin, then allow Automation, Contacts, and Full Disk Access
• Start a ChatGPT Work chat and prompt: “Check my unread iMessages and let me know what I need to catch up on. Then text myself the recap.” Make sure you have your own saved as a contact
• ChatGPT will ask you for read access to each unread conversation. Use “for this query” or approve each conversation globally for the rest of the ChatGPT session
• You should receive a text from ChatGPT summarizing your inbox!

Pro tip: The iMessage plugin is only available in the desktop app. However, you can click "Remote" in the iOS app to pair with your desktop or set a scheduled text triage task. For more info on these workarounds, watch the full video guide here.PRESENTED BY SONAR Deterministic rules meet AI-powered fixesThe Rundown: SonarQube provides strict standards to catch security vulnerabilities and tech debt across 40+ languages. Gitar’s AI complements this by reading your code context, identifying intent, analyzing root causes, and committing verified fixes directly to your branch.Gitar’s AI Code Review:

• Identifies root causes, applies fixes, and keeps the deployment pipeline flowing
• Saves senior engineer time on manual PR cycles
• Gives full-coverage, automated verification across every repository

Try them today.AI RESEARCH Anthropic tests automated AI safety researchImage source: AnthropicThe Rundown: Anthropic published new research where teams of Claude agents ran safety research on their own, successfully training away 10 kinds of AI misbehavior with results coming in over 4x better on average than human experts handed the same job.The details:

• Claude worked on 10 common failure types, including sycophancy, deception, and jailbreaks, cycling through literature searches, training, and scoring.
• The fixes ranged from a 26% decrease in sycophancy to a 96% improvement for reward hacking, without lowering overall abilities.
• Six veteran safety researchers saw just a 20% fix on deception with the same conditions, while Claude averaged 85% across over 150 attempts.
• A weaker Claude Sonnet 5 also spent 60 hours safety-training a pre-release Opus 4.8 build, using 15,000x less data than Anthropic's own process

Why it matters: Most AI roadmaps now assume the tech will eventually take over AI research, and this is one of the first concrete signs of the handoff actually starting. But with the fallout from OAI's breach still fresh and continuing to be uncovered, automating the safety side means placing a whole lot of trust in the non-human systems.QUICK HITSCOMMUNITY AI WORKFLOW OF THE DAY▸ Christian built a meal tracker he texts photos toToday’s workflow comes from reader Christian:“I built a meal tracker where my only job is to photograph a plate and send it to an agent. Identification, portion estimates, macros, storage, corrections, and the weekly review happen without me.The agent has its own email address and iMessage thread, so anything I can send from my phone becomes an input. I send corrections as plain English with 'CORRECTION' in the subject, and answer questions in the same thread. There's no app and no form.The accuracy mechanism is a challenge step. Every photo is estimated twice: first by the main model, then by a blind subagent that receives only the photo and the rubric — not the first answer. If it identifies different food, the entry is downgraded to low confidence and becomes a question for me.”See Christian’s workflow here. How do you use AI? Tell us for a chance to be featured. Trending AI Tools

•  Fireworks - Train an open model to be better, faster, cheaper on your work*
•  OpenClaw 2.0 - Open-source personal agent, now with multiplayer sessions
• Muse Code - Meta’s terminal-based AI coding agent, now out of beta
• Hermes Agent - Nous' Pantheon update turns its agents into a group chat

*Sponsored Listing Everything else in AI todaySpaceX’s Starshield AI rolled out Grok for Government on the Pentagon’s internal platform, opening the model for non-classified military use across its 3M employees. OpenAI announced that ChatGPT Ads crossed a $1B annualized run rate just 200 days after launch, with self-serve ad buying opening today across global markets.Build American AI, backed by a super PAC funded by Marc Andreessen and OAI's Greg Brockman, launched a multimillion-dollar ad campaign defending data centers. The EU designated ChatGPT as a “Very Large Online Platform”, requiring a ‘higher standard of scrutiny and accountability’ that OpenAI has four months to comply with. Bank of England Governor Andrew Bailey published a letter warning that frontier AI is showing “increasingly sophisticated autonomy and problem-solving abilities,” saying that the financial system lacks protocols to manage them. Highlights: News, Guides & Events

• Read our last AI newsletter: OpenAI cuts out SpaceX-owned Cursor
• Read our last Tech newsletter: Inside Meta’s $18B teen crackdown
• Read our last Robotics newsletter: Hugging Face’s robot duck is already a hit
• Today’s AI tool guide: How to connect ChatGPT to iMessage
• RSVP to next workshop on Sept. 2: Build an AI OS for your employee 1-on-1s

See you soon,Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown

Source: https://therundownai.beehiiv.com/p/runw ... e-internet