Samwise Tech/AI/Robotics Newsletter
Wednesday, May 13, 2026
Thinking Machines Lab Unveils Real-Time “Interaction Models” That Listen While They Speak
Mira Murati’s Thinking Machines Lab on May 12 unveiled a new class of AI system called “interaction models” — native multimodal systems designed to listen, speak, and see simultaneously without the turn-taking delays common in current voice AI. The company’s first release, TML-Interaction-Small, is a 276-billion-parameter mixture-of-experts model with 12 billion active parameters that achieves a conversational latency of 0.40 seconds, matching the pace of natural human speech. Standard APIs in 2026 typically introduce 1–2 second gaps. On Thinking Machines’ own benchmarks, the model scores an order of magnitude higher than OpenAI’s GPT Realtime-2. A broader rollout is expected later this year.
Sources: VentureBeat
Google Unveils Googlebooks: AI-Native Laptops Built Around Gemini to Succeed Chromebooks
Google unveiled Googlebooks on May 12 at its Android Show: I/O Edition event, announcing a new line of AI-native laptops built around Gemini, its flagship model family. Designed to succeed the Chromebook, Googlebooks run Android and feature a new AI-powered cursor called Magic Pointer that surfaces contextual suggestions based on whatever is on screen. Partners including Acer, Asus, Dell, HP, and Lenovo will manufacture the devices. Gemini intelligence is integrated at the OS level rather than added as an application. The company says Googlebooks will launch in fall 2026, though no pricing has been disclosed. The move signals Google’s shift away from ChromeOS after 15 years.
Sources: TechCrunch
Medicare’s New ACCESS Program Creates the First Federal Payment Model Built for AI-Driven Healthcare
The Centers for Medicare and Medicaid Services on May 12 launched ACCESS — Advancing Chronic Care with Effective, Scalable Solutions — a 10-year federal program built explicitly for AI-driven healthcare delivery. Unlike traditional Medicare, ACCESS pays participating organizations for patient health outcomes rather than activity counts, creating economics that favour AI-first operations. CMS selected 150 participants, including startup Pair Team, which has access to approximately 500,000 potential patients and is targeting one million within three years. The program goes live July 5. Critics note that patients are sharing highly sensitive health data through federal infrastructure with a documented history of security breaches.
Sources: TechCrunch
Waymo Issues Recall for 4,000 Robotaxis After Autonomous Vehicle Is Swept Away in Texas Flood
Waymo issued a software recall on May 12 covering nearly 4,000 autonomous robotaxis after vehicles struggled to navigate flooded roads in central Texas. In one incident, an empty robotaxi was swept away by floodwaters in San Antonio, prompting the company to pause operations in that city. The National Highway Traffic Safety Administration announced the recall, which deploys a software update restricting vehicle access to areas with elevated flood risk on high-speed roadways. Waymo says it identified a specific gap in its handling of flash flooding at highway speeds and is developing additional safeguards, including refined operational limits during periods of intense rain and geofencing for flood-prone areas.
Sources: TechCrunch
Exaforce Raises $125 Million at $725 Million Valuation to Detect and Stop Cyberattacks in Real Time
AI cybersecurity startup Exaforce raised a $125 million Series B on May 12, valuing the three-year-old company at $725 million. The round was led by Peak XV Partners with participation from HarbourVest, Mayfield, Khosla Ventures, and Seligman Ventures, bringing Exaforce’s total funding to $200 million. The company builds AI systems that detect and block cyberattacks in real time as they unfold, rather than after the fact. The raise arrives as AI-powered attacks accelerate sharply: 88 percent of enterprises reported AI-related security incidents in the past 12 months, and adversaries are using AI to exploit software vulnerabilities at unprecedented speed, according to CrowdStrike’s 2026 Global Threat Report.
Sources: TechCrunch
Google and SpaceX in Talks to Launch Orbital AI Data Centers Ahead of SpaceX IPO
Google and SpaceX are in early talks to put AI compute infrastructure in Earth orbit, The Wall Street Journal reported on May 12. Google is simultaneously negotiating with other launch providers, and its internal Project Suncatcher aims to launch prototype satellites by 2027. The discussions come as SpaceX prepares for its $1.75 trillion IPO and pitches investors on orbital compute as the cheapest long-term home for AI workloads. Anthropic last week signed a deal to use SpaceX’s xAI Memphis data center with an option on future orbital resources. Analysts caution that today’s terrestrial data centers remain substantially cheaper than orbital alternatives once satellite construction and launch costs are factored in.
Sources: TechCrunch
Altman Testifies Musk Wanted OpenAI Passed to His Children; Trial Sharpens Over For-Profit Conversion
OpenAI CEO Sam Altman took the stand on May 12 in Elon Musk’s lawsuit challenging the company’s for-profit conversion, delivering testimony that directly contradicted Musk’s claims about early organizational intent. Altman described a moment when Musk was asked what would happen if he died while controlling OpenAI; Musk reportedly said the organization should pass to his children. Altman said Musk’s management style had damaged research culture, noting that Musk required staff to rank colleagues by accomplishment — a practice Altman said caused long-lasting organizational harm. Musk previously testified that his company xAI trained its Grok model on OpenAI outputs. The trial is ongoing.
Sources: TechCrunch
Tech Pulse
Top Frontier Models (HLE): GPT-5.5 (54%) | Claude Opus 4.6 (51%) | Gemini 3.1 Pro (50%)
Top Open Source Models (SWE-Bench): Qwen3-235B-A22B (71%) | DeepSeek V4-Pro-Max (68%) | Mistral Large 3 (62%)
Top Small Models (15–50B): Gemma 3 27B (65%) | Qwen3-32B (64%) | Mistral Small 3.2 (59%)
Top Edge Models (0–15B): Phi-4 Mini (61%) | Gemma 3 12B (58%) | Mistral 3 7B (54%)
AI Leaders: NVIDIA $5.37T | Alphabet $4.8T | Microsoft $3.2T
Robotics Leaders: Tesla $1.5T | Intuitive Surgical $175B | ABB $165B
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.