- The Shift AI
- Posts
- Thinking Machines Drops Interactive AI
Thinking Machines Drops Interactive AI
Plus, ✨ How to edit, sign, and convert PDFs for free with RevPDF, Two new platforms reveal where AI competition is heading next, and more!
The Shift Newsletter
May 12, 2026
Sponsored by
Welcome back to The Shift. Let’s get straight to what matters in AI today…
Today we have:
🎙️ Mira Murati Just Launched an AI That Listens and Talks at the Same Time
✨ How to Edit, Sign, and Convert PDFs for Free With RevPDF
🤖 Two AI Infrastructure Platforms Launched This Week and Both Are Worth Watching
🔨Tools and Shifts you cannot miss
🎙️ Mira Murati Just Launched an AI That Listens and Talks at the Same Time
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released a research preview of its Interaction Model, an AI that processes audio, video, and text in parallel rather than waiting for you to finish speaking.
The Shift:
Every Voice AI Has a Latency Problem - Current voice AI listens, waits, processes, then responds, each step blocking the next. That adds up to 500ms plus of noticeable delay that breaks natural conversation flow entirely.
TML Processes in 200ms Chunks - The model listens, reasons, and speaks simultaneously in 200-millisecond windows. Average turn-taking latency hit 0.40 seconds on TML's benchmark, comparable to natural human conversation speed.
The Architecture Behind It - TML-Interaction-Small uses 276 billion total parameters but only 12 billion active per inference. A separate Background Agent handles deeper reasoning in parallel while the Interaction Model keeps the conversation flowing.
Where It Falls Short - With only 12 billion active parameters, reasoning quality is good but not exceptional. GPT-4 and Claude still beat it on pure capability and benchmarks are on TML's own internal dataset.
Research preview is live now for selected researchers with no API yet. API access for partners is planned for Q2 2026, open beta Q3 2026, and production release by Q4 2026. TML is also funding $5 million plus in AI safety grants.
Together with Readdy.ai
Your Business Website. Built in Under 2 Minutes
Running a business is hard. Building a website shouldn't be.
With Readdy.ai, just describe your business, the AI handles your website design, performance, SEO optimization, and more.
No coding or design skills needed. Launch in 2 minutes. Your site could be live before you finish your coffee.
✨ How to Edit, Sign, and Convert PDFs for Free With RevPDF
Everything Adobe charges for. 20MB app. Runs entirely offline.
Step 1: Download RevPDF - Head to revpdf.com. No account, no subscription, no internet required after install.
Step 2: Edit text and images directly - Open any PDF and edit content in place, modify fonts, reposition elements, update text with precision.
Step 3: Redact sensitive information - Permanently remove confidential text and images before sharing. Clean and irreversible.
Step 4: Fill and sign forms - Complete any PDF form directly, fields, checkboxes, dropdowns. Draw, type, or upload your signature and save it for reuse.
Step 5: Compress, split, or merge - Shrink file size without visible quality loss, split by page range, or combine multiple documents into one.
Step 6: Convert and export - Convert PDF pages to images, images to PDF, or export to Word, all offline.
No cloud. No data is leaving your device. No monthly fee. You can download RevPDF free here.
🤖 Two AI Infrastructure Platforms Launched This Week, and Both Are Worth Watching
One lets you explore AI-generated worlds in real time from your browser. The other lets anyone build and deploy custom frontier models. Different products, same underlying bet, infrastructure is where the next AI race is won.
The Shift:
Reactor launched its world generation platform this week, letting anyone explore AI-generated environments that render live directly from a browser, no downloads, no setup. Watch the video here.
The lab is positioning itself as an infrastructure layer, betting that real-time delivery will become the category's main bottleneck as world models scale. The launch video hit 7.5 million views.
Prime Intellect's Lab came out of beta this week, giving anyone the ability to build, evaluate, and deploy custom frontier models and self-improving agents trained on their own data.
The platform bundles everything needed to run a model and benchmark its performance in one place, removing the infrastructure overhead that previously made custom model development inaccessible outside well-funded labs. Watch it in action
Both launches point to the same shift: the hardest part of AI is no longer the model itself but the infrastructure around it. Reactor is betting on delivery. Prime Intellect is betting on training. Both are making previously gated capabilities accessible to anyone.
Together with The Marketing Millennials
The best marketing ideas come from marketers who live it. That’s what The Marketing Millennials delivers: real insights, fresh takes, and no fluff. Written by Daniel Murray, a marketer who knows what works, this newsletter cuts through the noise so you can stop guessing and start winning. Subscribe and level up your marketing game.
🔨AI Tools for the Shift
🎸 TonesMatch – Generates exact amp settings, pedal chains, and pickup configurations for songs based on your specific guitar and gear setup.
📊 GanttPRO – Instantly creates AI-powered project timelines to help teams visualize, organize, and manage work faster.
📄 PDF Generator by PDF.net – Turns simple prompts into professional PDFs like invoices, resumes, agreements, and forms without using templates.
🪞 Mirror Test – Creates a surprisingly accurate personality reflection by analyzing answers from people close to you.
🧭 PathSpotter – Helps people make clearer decisions with guided insights designed for those tired of guessing their next move.
🚀Quick Shifts
🏠 Google says Gemini for Home is getting faster smart home controls through backend optimizations, while updated age and content controls expand responses for requests like cocktail recipes and other restricted content.
🧠 OpenAI co-founder Ilya Sutskever says he was uncomfortable with Elon Musk’s demands for large ownership and board control, arguing that merging OpenAI into Tesla “would kill a dream” for the company’s future.
⚖️ OpenAI faces a lawsuit from the family of an FSU shooting victim, alleging ChatGPT’s design assisted the accused shooter, while OpenAI says the chatbot only provided publicly available factual information.
⚖️ Microsoft CEO Satya Nadella says OpenAI’s board appeared “sort of amateur city” after Sam Altman’s firing, with Microsoft receiving little explanation despite the significance of removing a CEO.
🏢 OpenAI launches the OpenAI Deployment Company, a new enterprise-focused division designed to help organizations build, deploy, and scale AI systems through embedded “Forward Deployed Engineers” working directly inside businesses.
🧩 Prompt of the Day
How-to Content That Educates and Builds Trust
Most tutorials overwhelm users or stay too surface-level. Great how-to content simplifies the process, removes confusion, and makes the audience feel capable immediately.
Turn education into engagement and authority.
Paste the prompt: Drop this into ChatGPT, then fill in your topic details.
Prompt to paste
Create a detailed how-to guide for [Insert topic or product]. Break the tutorial into clear step-by-step instructions that are easy for beginners to follow. Suggest visual aids such as screenshots, diagrams, short videos, or GIFs that improve understanding at each stage. Include practical tips, common mistakes to avoid, and a short summary of the expected outcome. Focus on making the content educational, actionable, and highly engaging.
That’s all for today’s edition. See you tomorrow as we track down and get you all that matters in the daily AI Shift!
If you loved this edition, let us know how much:
How useful was today's edition? |
Forward it to your pal to give them a daily dose of the shift so they can 👇
Reply