Samwise Tech/AI/Robotics Newsletter
Friday, April 24, 2026
OpenAI Releases GPT-5.5, Its Most Capable Model Yet
OpenAI released GPT-5.5 on Thursday, describing it as the company’s smartest and most intuitive model to date. Chief Research Officer Mark Chen said the model delivers meaningful gains on scientific and technical research workflows, including drug discovery and mathematics. GPT-5.5 consistently outperforms Gemini 3.1 Pro and Claude Opus 4.5 on major benchmarks, according to OpenAI. The model is now available to Plus, Pro, Business, and Enterprise users in ChatGPT, with the more capable GPT-5.5 Pro rolling out to higher-tier subscribers. OpenAI positions the release as another step toward a unified AI super app combining chat, reasoning, and agentic capabilities.
Sources: TechCrunch
Google Cloud Launches Two New AI Chips to Compete With Nvidia
Google Cloud unveiled two new AI chip families at its Cloud Next 2026 conference on April 22: the TPU 8t for model training and the TPU 8i for inference. The TPU 8t delivers nearly 3x higher compute performance than its predecessor, packing 9,600 chips into a single superpod providing 121 exaflops of compute and two petabytes of shared memory. The TPU 8i is optimized for ultra-low latency required by agentic workflows, carrying 384 megabytes of on-chip SRAM. Both chips deliver up to 3x faster training and 2.8x better price-to-performance compared to prior generations. General availability is expected later in 2026.
Sources: TechCrunch
Anthropic Launches Project Glasswing to Deploy Restricted AI Model Against Zero-Day Vulnerabilities
Anthropic has launched Project Glasswing, a cybersecurity initiative pairing its unreleased Claude Mythos Preview model with a coalition of major technology companies including Microsoft, Amazon, Google, and Apple to identify vulnerabilities in critical software infrastructure. Anthropic says Mythos Preview has already detected thousands of high-severity zero-day vulnerabilities across major operating systems and web browsers, including a 27-year-old bug in OpenBSD. Because of the model’s exceptional offensive capabilities, Anthropic is not making it publicly available. Microsoft is providing Azure customers access to Mythos Preview through Project Glasswing for defensive security work under controlled conditions.
Sources: VentureBeat
Google Signs Multibillion-Dollar Cloud Deal With Mira Murati’s Thinking Machines Lab
Google has deepened its relationship with Mira Murati’s Thinking Machines Lab through a new multibillion-dollar cloud infrastructure deal, Thinking Machines’ first partnership with a cloud provider. Under the agreement, the startup gains access to Google Cloud infrastructure powered by Nvidia’s GB300 chips, which offer 2x improvements in training and serving speed over prior-generation GPUs. The deal, valued in the single-digit billions, is non-exclusive, allowing Thinking Machines to use multiple cloud providers. The lab is among the first Google Cloud customers to access GB300-powered systems and will use them to train and deploy its AI models.
Sources: TechCrunch
Google Turns Chrome Into an AI Co-Worker Capable of Autonomous Web Tasks
Google announced plans to bring agentic capabilities to Chrome for enterprise users, turning the browser into an AI assistant capable of taking actions across websites on a user’s behalf. The feature uses Gemini to understand context across open browser tabs and automate tasks such as booking travel, entering data into CRM systems, comparing vendor pricing, and summarizing candidate portfolios. The tool is initially available to Workspace users in the United States and can be enabled via administrator policy, with Google stating that user prompts will not be used to train its models.
Sources: TechCrunch
AI Agents Score Half as Well as PhD Scientists on Complex Research Tasks, Nature Reports
A study published in Nature on April 23 found that AI agents still perform significantly worse than human scientists with PhDs on complex, multi-step research tasks. The Stanford AI Index 2026, cited in the report, showed that leading frontier AI agents score roughly half as well as expert human researchers on demanding scientific benchmarks. Despite growing adoption of AI tools in laboratories — the number of scientific publications mentioning AI grew nearly 30-fold from 2010 to 2025 — autonomous AI agents remain unable to replicate the adaptive reasoning that trained scientists routinely apply to novel problems.
Sources: Nature
Humble Robotics Exits Stealth With Cabless Autonomous Electric Hauler and $24M in Funding
Humble Robotics emerged from stealth on April 23 with its autonomous, cabless electric hauler designed for logistics environments including warehouses, railyards, and seaports. The company simultaneously disclosed $24 million in seed funding led by Eclipse with participation from Energy Impact Partners. Humble’s vehicle is a reimagined Class 8 freight system significantly lighter than a traditional tractor-trailer combination by eliminating the cab, enabling new logistics use cases. The hauler uses camera, LiDAR, and radar sensors in a 360-degree configuration for dock-to-dock autonomous operation. The first model is built specifically to move shipping containers.
Sources: Robotics & Automation News
Tech Pulse
Top Models: Claude Mythos Preview (93.9%) | Claude Opus 4.7 (87.6%) | GPT-5.3 Codex (85.0%)
Market: NVIDIA $4.60T | Alphabet $3.89T | Apple $3.81T
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.