Large Language Models (LLMs) – Overview and Latest News

BitNet b1.58 2B4T from Microsoft Research aims for efficient AI use on CPUs with a native 1.58-bit architecture and custom frameworks.

To combat server strain from AI bots, Wikimedia Enterprise has made a structured Wikipedia dataset available via Google's Kaggle platform.

Microsoft has integrated OpenAI's new o3 and o4-mini reasoning models, featuring agentic capabilities and vision, into Azure OpenAI Service and GitHub Copilot.

Meta has blocked Apple Intelligence features like Writing Tools and Genmoji from working in its iOS apps, including Facebook, Instagram, and WhatsApp.

Elon Musk's xAI has launched a Memory feature for its Grok chatbot, enabling personalized recall of past conversation details across sessions.

Perplexity AI has reportedly entered talks with Samsung to feature its assistant on Galaxy phones, providing a potential alternative to Google Gemini following a deal with Motorola.

OpenAI has entered talks to acquire Windsurf for $3B, expanding its developer tool strategy and challenging GitHub Copilot in AI code generation.

Codex CLI is a free, open source tool for model-agnostic AI coding that lets developers use AI directly in the terminal without relying on cloud-based platforms.

OpenAI has launched o3 and o4-mini, giving ChatGPT the ability to reason and choose tools autonomously across text, code, image, and file inputs.

Paying Claude users can now integrate the AI with their Google Workspace apps and use a new agentic Research feature combining internal context with web search.

OpenAI has updated its safety framework, introducing a clause allowing rules to be adjusted based on competitor actions amid scrutiny over testing speed.

OpenAI is working on a social feed prototype using ChatGPT image generation, aiming to compete with X and Meta while gathering data to train future AI models.

Ilya Sutskever's AI startup SSI has secured $2B from Alphabet, Nvidia, and VCs, reaching a $32B valuation and partnering with Google Cloud for TPU use.

Google is working on DolphinGemma, a new AI based on its Gemma architecture, which analyzes decades of dolphin recordings from the Wild Dolphin Project.

OpenAI has outlined plans for potential government ID verification for API users seeking access to advanced future models like o3 and GPT-4.1.

OpenAI has significantly reduced safety testing time for new AI models like o3, raising alarms among testers about rushing powerful tech due to competitive pressures.

ChatGPT's improved memory capability has launched for paid users, enabling recall of conversation history for improved relevance, with user controls available.

Meta's new Llama 4 AI models have been developed to address perceived left-leaning bias, aiming for political balance alongside platform policy shifts.

OpenAI is set to launch GPT-4.1 and reasoning models o3/o4-mini soon, reversing earlier plans and delaying GPT-5 amidst capacity issues.

Google Workspace has received major AI updates, including the new Flows automation platform and enhanced Gemini capabilities within core productivity apps.

With its latest AI model, Google pivots toward throughput over power, offering businesses a lighter, faster alternative to its Pro-tier systems—but not without raising concerns around safety and transparency.

OpenAI and Google have opposed the UK’s proposal for an AI copyright watchdog, warning it could introduce legal uncertainty and stifle innovation.

DeepSeek has unveiled SPCT, a self-critique-based AI tuning method that outperforms traditional alignment techniques without increasing model size.

OpenAI is testing a visible watermark for mages created with ChatGPT-4o i as part of its broader efforts to improve AI content traceability and policy compliance.

Meta has launched Llama 4 Scout and Maverick with image-text fusion, new architecture, and early results rivaling GPT-4o and Gemini 2.0 in benchmarks.

OpenAI has restructured its roadmap to decouple reasoning from chat models, starting with the release of o3 and o4-mini in the coming weeks.

GitHub has launched Agent Mode for Copilot, enhanced AI model access wit a new Pro+ plan, and rolled out AI powered code reviews.

Microsoft has launched new Copilot tools including Pages, Shopping, and Copilot Search, and has removed Pro-only restrictions for key features

Anthropic has launched Claude for Education to help universities adopt transparent, critical-thinking-focused AI across classrooms and campuses.

Meta’s top AI researcher Joëlle Pineau will leave the company in May 2025, marking a significant shift as the company faces regulatory scrutiny and internal challenges.