MIT Technology Review examines why AI adoption in agriculture has underperformed relative to its potential, identifying fragmented and poor-quality data infrastructure as the primary bottleneck. While agricultural use cases—crop yield prediction, pest identification, soil optimization—are theoretically well-suited for AI, most farms lack standardized, centralized data collection systems needed to train effective models.
Anthropic has launched Claude Sonnet 5, a new version of its popular AI model designed to deliver better performance at lower operating costs. The update addresses a key pain point for businesses deploying AI agents and automation workflows—the expense of running advanced models at scale. Claude Sonnet 5 maintains the reasoning and coding capabilities of its predecessor while reducing computational overhead, making it a more practical choice for enterprises managing large deployments.
EquiLibre Technologies, a Prague-based AI laboratory founded by three former DeepMind researchers, has reached a $500 million valuation by applying machine learning techniques to quantitative trading and financial modeling. The team's expertise in game theory and strategic decision-making—developed while building poker AI systems—translates directly to modeling market behavior and identifying trading patterns.
Etched, an AI chip startup positioning itself as an alternative to NVIDIA, has announced a $5 billion valuation and $1 billion in contracted orders for its inference hardware systems. The company's architecture focuses specifically on optimizing the compute requirements for running large language models efficiently, rather than building general-purpose AI chips.
Flexion Robotics, founded by former NVIDIA engineers, has developed a humanoid robot capable of performing practical office and light industrial tasks. The company uses a training methodology that enables robots to learn useful work patterns more efficiently than traditional robotics approaches, suggesting a path toward commercially viable humanoid automation.
Google is updating its image generation capabilities with Nano Banana 2 Lite, an optimized version of its imaging model that delivers faster processing times and reduced costs per generated image. The update targets creative teams and marketing departments that rely on AI-generated visuals for campaigns, website content, and digital assets.
Google is adding the ability to automatically generate 60-second vertical video clips directly from notes and uploaded research documents in NotebookLM. The feature, rolling out to Google AI Ultra and Pro subscribers, transforms static notes into social media-ready content without manual editing or production work.
WIRED reports that hundreds of contractors working for Meta posed as teenagers to test how rival chatbots—including Google's Gemini and OpenAI's ChatGPT—would respond to high-risk queries about suicide, drug use, and sexual content. The testing program, spanning multiple months, was designed to identify safety gaps and competitive vulnerabilities in competing AI systems.
OpenAI has published new data from its Signals platform showing how ChatGPT adoption is expanding globally, with increases in user engagement, exploration of advanced features, and growth across different regions and language groups. The data reflects broader trends in where enterprises are finding value from conversational AI.
OpenAI has released GeneBench-Pro, a benchmarking tool that measures AI model performance on genomics, biology, and scientific research tasks using real-world datasets. The benchmark goes beyond synthetic test cases to evaluate how AI systems handle the complexity and noise present in actual research data.