🔍 Google Search will become fully conversational with Gemini 3 🔍
New experience will let users seamlessly shift from search to AI dialogue

Google announced a fundamentally new Search experience coming soon that will transform how users interact with information online. With Gemini 3 set to power AI Overviews by default, users will be able to seamlessly transition from traditional search results into full conversational AI Mode without losing context. The redesigned interface will offer "just ask anything" prompts directly within search results, letting users explore topics through natural back-and-forth dialogue. Google describes the future experience as delivering a quick snapshot when you need it, deeper conversation when you want it—marking Search's evolution from link aggregator to AI conversation partner.

🔬 OpenAI releases Prism for scientific research 🔬
GPT-5.2-powered LaTeX editor accelerates paper writing

OpenAI released Prism, a free AI-powered workspace designed to revolutionize scientific research. The tool enables researchers to draft papers, manage citations, summarize literature, and convert whiteboard photos into formatted equations and diagrams. With ChatGPT already receiving 8.4 million weekly queries on advanced science and math topics, Prism represents OpenAI's strategic move into specialized professional tools. Available to all ChatGPT users at no cost.

🤖 Moonshot's Kimi K2.5 rivals GPT-5.2 🤖
Open-source model coordinates up to 100 parallel agents

Chinese AI startup Moonshot AI released Kimi K2.5, a natively multimodal open-source model matching leading proprietary models across coding and vision benchmarks. The standout Agent Swarm feature coordinates up to 100 parallel sub-agents across 1,500+ tool calls using Parallel-Agent Reinforcement Learning (PARL). This reduces execution time by up to 80% compared to single-agent setups. Built with 15 trillion mixed visual and text tokens, K2.5 delivers state-of-the-art performance rivaling GPT-5.2 and Claude Opus 4.5.

INTERESTING TO KNOW

🧠 AI models can now teach themselves 🧠

"Recursive self-improvement" is a fast-growing research focus drawing interest from startups and leading labs including Google DeepMind. DeepMind CEO Demis Hassabis confirmed the company is exploring whether models can "continue to learn out in the wild after you finish training them." A new Georgetown CSET report warns that while AI systems accelerating their own progress could sustain rapid advancement, they also introduce new risks harder to detect and control. OpenAI CEO Sam Altman aims to build a "true automated AI researcher" by March 2028.

🚗 Waabi raises $1B and partners with Uber for robotaxis 🚗

Nvidia has invested $2 billion in CoreWeave to accelerate the buildout of over 5 gigawatts of AI factory capacity by 2030. This infrastructure investment signals intensifying competition in AI compute as enterprise demand surges, deepening the strategic partnership between the leading GPU maker and one of the fastest-growing AI cloud providers.