Apple has ceased signing iOS 18.0.1 a week after releasing iOS 18.1, preventing iPhone and iPad users from downgrading. iOS 18.1 introduced new smart tools and security enhancements, including fixes for a Siri vulnerability. Users on iOS 17 can continue receiving security updates without upgrading, but cannot revert once on iOS 18.

Google Research launched InkSight, an AI-powered system that directly converts handwritten photos to digital text, outperforming OCR in accuracy and reliability even under challenging conditions. This breakthrough advances handwriting recognition and promises broad application.

Anthropic launches Haiku 3.5 AI model, integrating with major platforms. Excelling in coding and reasoning, it’s priced higher than its predecessor. Optimizations reduce costs. Upgraded Claude 3.5 Sonnet outperforms peers in programming tasks, showcasing Anthropic’s AI expertise and commitment to advancing technology.

Physical Intelligence, a robotics startup led by ex-Google scientist Carol Hausman, raised $400M, reaching a $2.4B valuation. Backed by Jeff Bezos, Thrive Capital, and OpenAI, it develops universal AI models enabling robots to perform complex, adaptable tasks in dynamic environments.

Amazon Prime Video introduces X-Ray Recaps, allowing users to swiftly review seasons, episodes, or segments with text summaries. Currently testing on Fire TV, it will expand to more devices and Amazon MGM Studios originals by year-end.

Former OpenAI policy head Miles Brundage predicts AGI could perform most human computer tasks and emerge by 2026. Leaving OpenAI to influence AI policy and regulation, amid industry scrutiny over AGI development and safety.

xAI, backed by Elon Musk, launched its API offering $25/month free credits to developers until year-end. Leveraging X data, Grok LLMs are competitively priced. The API supports 128k tokens and is compatible with OpenAI and Anthropic SDKs. New Grok models and the Colossus supercluster are forthcoming.

Apple integrates ChatGPT into iOS 18.2, offering ChatGPT Plus for $20/month via Settings, potentially boosting OpenAI’s revenue. Free users face limitations. Monetization details are unclear. Apple may add other AI models. OpenAI faces funding issues as Apple withdrew a $6.6B investment after the CTO's resignation.

Kathleen Carlinovskaya, ex-Meta AR/VR leader, joins OpenAI to head Robotics and Consumer Hardware. She may collaborate with Jony Ive’s LoveFrom on AI devices. OpenAI revives its robotics division, partnering with Apple and Figure on AI-integrated hardware.

Professor Zhou Yongfeng's team published in Nature Genetics, using AI to advance grape breeding. They developed Grapepan v1.0, identified 148 trait loci, and created a machine learning model with 85% accuracy, enhancing breeding efficiency and innovation. Secured six national and one international patents.

AWS CEO Matt Garman warns that training future large language models may consume energy comparable to major cities. To address rising demands, AWS invests in renewable and nuclear projects, highlighting energy supply as a key factor for AI advancement.

ChatGPT Search revolutionizes online searches with conversational AI, offering concise answers and real-time data. Sam Altman highlights AI's economic impact and introduces Simple Bench for AI benchmarking. Future trends include enhanced integration, advanced NLP, and greater transparency.

NVIDIA may invest in Elon Musk's AI startup xAI, targeting a $75B valuation with a January funding round. xAI raised $6B, launched a top AI cluster, and plans expansion. Additionally, MIT's Liquid AI develops efficient neural networks to reduce costs and enhance AI adoption.

OpenAI CEO Sam Altman announced on Reddit AMA that limited computational capacity is delaying new product launches like ChatGPT’s Advanced Voice Mode and DALL-E updates. Despite challenges, OpenAI continues innovating with web-integrated ChatGPT and explores AI’s role in shopping.

Researchers developed a framework to train smaller LLMs with function calls for logic and math reasoning. Using proxy programs and fine-tuning, this approach enhances task performance while reducing computational costs, demonstrating significant accuracy improvements in experiments.

Meta AI introduces Sparsh, a universal vision-based tactile encoder utilizing self-supervised learning. Sparsh addresses sensor diversity and minimizes labeled data requirements, enhancing robotic interactions and achieving 95% performance on TacBench benchmarks.

MIT launches the Heterogeneous Pre-training Transformer, a robot training framework using diverse sensor data akin to large language models. It aims to create a universal robotic "brain" adaptable without extra training. Funded by Toyota and partnering with Boston Dynamics.

Hugging Face launches SmolLM2, a compact language model series (135M-1.7B parameters) optimized for edge devices. The 1.7B variant outperforms Meta's Llama 1B in key benchmarks. Licensed under Apache 2.0, it offers efficient AI capabilities for local deployment, supporting tasks like summarization and function calling.

Google's Gemini AI now auto-summarizes Google Chat conversations via the sidebar, enhancing efficiency and managing information. It helps review chats and create to-do lists, though users should verify critical points since nuances may be missed. This feature showcases AI's growing workplace role, urging careful use to maintain quality.

Google launches "Learn About" at I/O, an experimental AI tool offering personalized, conversational learning by integrating generative AI with educational content. Acting as a digital tutor, it transforms research and study. Currently testing, Google seeks user feedback to improve this AI-driven learning companion.