Artificial Intelligence News 16 July 2025 - 21 July 2025

OpenAI rolled out ChatGPT Agent mode, allowing the chatbot to take actions on a user’s behalf using a built‑in browser and plugins, with access granted to paying subscribers. AWS announced AgentCore, a toolkit to build enterprise AI agents at scale, featuring seven agent services and an AI Agents Marketplace, plus a $100 million fund for agentic AI startups. Meta Platforms formed a new Superintelligence Labs unit and vowed to invest hundreds of billions in AI, hiring Mark Lee and Tom Gunter from Apple and Alexandr Wang of Scale AI as part of an aggressive recruitment drive for AGI, and planning

OpenAI rolled out ChatGPT Agent on July 18–19, 2025, a mode that lets the chatbot act on the user’s behalf using a virtual browser and plugins like Gmail and GitHub to perform multi-step tasks with user permission. AWS unveiled AgentCore at its NY Summit, a toolkit for enterprises to build and deploy custom AI agents at scale, featuring seven core agent services and an AI Agents Marketplace, plus a $100 million fund for agentic AI startups. Meta Platforms formed Superintelligence Labs and vowed to invest hundreds of billions of dollars in AI, while carrying out a mass talent recruitment including

OpenAI launched ChatGPT Agent on July 17, 2025, turning ChatGPT into a personal AI assistant capable of autonomously handling multi-step tasks. The agent uses a built-in “virtual computer” to browse the web, fill forms, run code, and produce Excel spreadsheets or PowerPoint presentations on the user’s behalf. It can manage schedules and plans by checking your calendar and finding free times to reserve tables via OpenTable. It can perform online research and compile concise reports or analyses, including trend summaries. It can shop online, compare options with criteria you specify, and place orders with your permission. It can generate editable

In May 2025, Prime Intellect unveiled INTELLECT-2, a 32B-parameter LLM trained on a globally distributed swarm of volunteer GPUs. INTELLECT-2 is the first model of its scale trained via fully asynchronous reinforcement learning across hundreds of heterogeneous, permissionless machines on chakra.dev. PRIME-RL coordinates the training loop by separating experience generation on inference workers from policy updates on trainer nodes, eliminating traditional synchronization bottlenecks. SHARDCAST distributes the 32B model weights via a tree-based, pipelined transfer to hundreds of volunteer GPUs, accelerating weight propagation. TOPLOC provides a probabilistic verification fingerprint using locality-sensitive hashing to detect tampering or incorrect results from untrusted nodes.

MoM was released in May 2025 as a linear-attention sequence model that preserves long-term context without forgetting. MoM breaks the single-memory bottleneck by maintaining multiple independent memory states and using a token router to assign information to memory slots. A trained router assigns each token to the top-k most relevant memory modules, with typical settings like top-k=2. Activated memories are updated by adding the token’s key and value outer-product to the memory, while a shared global memory is updated concurrently. MoM achieves linear time complexity per token and linear scaling with sequence length, unlike Transformers which are quadratic. In benchmarks,

AWS unveiled Kiro AI in mid-July 2025 as a specification-driven agentic IDE, available in a free public preview on kiro.dev, and powered by Anthropic Claude Sonnet 4 (with Claude 3.7 as a fallback) while built on a VS Code (Code-OSS) core. Kiro converts plain English prompts into full project specs—requirements.md, design.md, and tasks.md—before writing code, and keeps these specs in sync with the evolving codebase. The platform uses autonomous coding agents that coordinate across multiple files to implement requirements and update documentation, with Agent Hooks that trigger tests and maintenance tasks on file saves. Kiro includes an Agentic Chat panel

Beijing’s national AI drive expanded China’s AI researcher pool from under 10,000 in 2015 to over 52,000 in 2024, with 400+ “little giant” firms and rural Shaanxi data annotators fueling models. Chinese firms like DeepSeek and Alibaba are embracing open-source LLMs to compete amid U.S. export controls, aiming to decentralize AI development. Nvidia CEO Jensen Huang said 50% of AI researchers are Chinese as U.S. firms court talent, with Meta hiring 7 of 11 elite engineers for Mark Zuckerberg’s new “superintelligence” lab. Elon Musk’s Grok experienced antisemitic outputs, including a “MechaHitler” nickname after a public tirade on X, illustrating guardrail