Zero-shot and few-shot learning are techniques that enable AI models to recognize and generalize to new data with minimal training. Instead of showing a model thousands of pictures of cats, zero-shot learning aims to have it identify a cat based purely on a description, while few-shot learning might need just a handful of examples.

These approaches are useful when labeled data is scarce or expensive to obtain. Imagine teaching an AI to identify different species of plants in a rainforest; collecting and labeling vast amounts of data would be incredibly difficult. 

For example, a language model might be able to translate English to a new language (zero-shot) or learn to play a new game (few-shot) with only a few demonstrations.

A key limitation is that performance often lags behind fully supervised models trained on massive datasets. Results can also be unpredictable!

Training AI models can take *a lot* of data. What if you could get similar results with far less?

AI 101: Zero-shot & Few-shot Learning

Accelerating discovery with the AI for Math Initiative

Google DeepMind is spearheading a new era of mathematical discovery with its AI for Math Initiative, designed to leverage artificial intelligence in solving intricate mathematical challenges. Partnering with leading institutions like Imperial College London and UC Berkeley, the initiative aims to identify mathematical problems ripe for AI insights, develop essential tools, and accelerate research. Google's support includes funding and access to advanced AI technologies like Gemini Deep Think and AlphaEvolve, which have already demonstrated remarkable abilities, such as achieving gold-medal level performance at the International Mathematical Olympiad and discovering more efficient methods for matrix multiplication. This initiative seeks to blend the intuition of mathematicians with AI's capabilities, potentially unlocking new scientific breakthroughs.

Google DeepMind's 'AI for Math' initiative unites top research institutions to explore AI's potential in solving complex mathematical problems.

AI Revolutionizes Mathematics: New Initiative Launches

Vibe coding platform Cursor releases first in-house LLM, Composer, promising 4X speed boost

The vibe coding tool Cursor, from startup Anysphere, has introduced Composer, its first in-house, proprietary coding large language model (LLM) as part of its Cursor 2.0 platform update. Composer is designed to execute coding tasks quickly and accurately in production-scale environments, representing a new step in AI-assisted programming. It's already being used by Cursor’s own engineering staff in day-to-day development — indicating maturity and stability. According to Cursor, Composer completes most interactions in less than 30 seconds while maintaining a high level of reasoning ability across large and complex codebases. The model is described as four times faster than similarly intelligent systems and is trained for “agentic” workflows—where autonomous coding agents plan, write, test, and review code collaboratively. Previously, Cursor supported "vibe coding" — using AI to write or complete code based on natural language instructions from a user, even someone untrained in development — atop other leading proprietary LLMs from the likes of OpenAI, Anthropic, Google, and xAI. These options are still available to users. Benchmark Results Composer’s capabilities are benchmarked using "Cursor Bench," an internal evaluation suite derived from real developer agent requests. The benchmark measures not just correctness, but also the model’s adherence to existing abstractions, style conventions, and engineering practices. On this benchmark, Composer achieves frontier-level coding intelligence while generating at 250 tokens per second — about twice as fast as leading fast-inference models and four times faster than comparable frontier systems. Cursor’s published comparison groups models into several categories: “Best Open” (e.g., Qwen Coder, GLM 4.6), “Fast Frontier” (Haiku 4.5, Gemini Flash 2.5), “Frontier 7/2025” (the strongest model available midyear), and “Best Frontier” (including GPT-5 and Claude Sonnet 4.5). Composer matches the intelligence of mid-frontier systems while delivering the highest recorded generation speed among all tested classes. A Model Built with Reinforcement Learning and Mixture-of-Experts Architecture Research scientist Sasha Rush of Cursor provided insight into the model’s development in posts on the social network X , describing Composer as a reinforcement-learned (RL) mixture-of-experts (MoE) model: “We used RL to train a big MoE model to be really good at real-world coding, and also very fast.” Rush explained that the team co-designed both Composer and the Cursor environment to allow the model to operate efficiently at production scale: “Unlike other ML systems, you can’t abstract much from the full-scale system. We co-designed this project and Cursor together in order to allow running the agent at the necessary scale.” Composer was trained on real software engineering tasks rather than static datasets. During training, the model operated inside full codebases using a suite of production tools—including file editing, semantic search, and terminal commands—to solve complex engineering problems. Each training iteration involved solving a concrete challenge, such as producing a code edit, drafting a plan, or generating a targeted explanation. The reinforcement loop optimized both correctness and efficiency. Composer learned to make effective tool choices, use parallelism, and avoid unnecessary or speculative responses. Over time, the model developed emergent behaviors such as running unit tests, fixing linter errors, and performing multi-step code searches autonomously. This design enables Composer to work within the same runtime context as the end-user, making it more aligned with real-world coding conditions—handling version control, dependency management, and iterative testing. From Prototype to Production Composer’s development followed an earlier internal prototype known as Cheetah , which Cursor used to explore low-latency inference for coding tasks. “Cheetah was the v0 of this model primarily to test speed,” Rush said on X. “Our metrics say it [Composer] is the same speed, but much, much smarter.” Cheetah’s success at reducing latency helped Cursor identify speed as a key factor in developer trust and usability. Composer maintains that responsiveness while significantly improving reasoning and task generalization. Developers who used Cheetah during early testing noted that its speed changed how they worked. One user commented that it was “so fast that I can stay in the loop when working with it.” Composer retains that speed but extends capability to multi-step coding, refactoring, and testing tasks. Integration with Cursor 2.0 Composer is fully integrated into Cursor 2.0, a major update to the company’s agentic development environment. The platform introduces a multi-agent interface, allowing up to eight agents to run in parallel, each in an isolated workspace using git worktrees or remote machines. Within this system, Composer can serve as one or more of those agents, performing tasks independently or collaboratively. Developers can compare multiple results from concurrent agent runs and select the best output. Cursor 2.0 also includes supporting features that enhance Composer’s effectiveness: In-Editor Browser (GA) – enables agents to run and test their code directly inside the IDE, forwarding DOM information to the model. Improved Code Review – aggregates diffs across multiple files for faster inspection of model-generated changes. Sandboxed Terminals (GA) – isolate agent-run shell commands for secure local execution. Voice Mode – adds speech-to-text controls for initiating or managing agent sessions. While these platform updates expand the overall Cursor experience, Composer is positioned as the technical core enabling fast, reliable agentic coding. Infrastructure and Training Systems To train Composer at scale, Cursor built a custom reinforcement learning infrastructure combining PyTorch and Ray for asynchronous training across thousands of NVIDIA GPUs. The team developed specialized MXFP8 MoE kernels and hybrid sharded data parallelism, enabling large-scale model updates with minimal communication overhead. This configuration allows Cursor to train models natively at low precision without requiring post-training quantization, improving both inference speed and efficiency. Composer’s training relied on hundreds of thousands of concurrent sandboxed environments—each a self-contained coding workspace—running in the cloud. The company adapted its Background Agents infrastructure to schedule these virtual machines dynamically, supporting the bursty nature of large RL runs. Enterprise Use Composer’s performance improvements are supported by infrastructure-level changes across Cursor’s code intelligence stack. The company has optimized its Language Server Protocols (LSPs) for faster diagnostics and navigation, especially in Python and TypeScript projects. These changes reduce latency when Composer interacts with large repositories or generates multi-file updates. Enterprise users gain administrative control over Composer and other agents through team rules, audit logs, and sandbox enforcement. Cursor’s Teams and Enterprise tiers also support pooled model usage, SAML/OIDC authentication, and analytics for monitoring agent performance across organizations. Pricing for individual users ranges from Free (Hobby) to Ultra ($200/month) tiers, with expanded usage limits for Pro+ and Ultra subscribers. Business pricing starts at $40 per user per month for Teams, with enterprise contracts offering custom usage and compliance options. Composer’s Role in the Evolving AI Coding Landscape Composer’s focus on speed, reinforcement learning, and integration with live coding workflows differentiates it from other AI development assistants such as GitHub Copilot or Replit’s Agent. Rather than serving as a passive suggestion engine, Composer is designed for continuous, agent-driven collaboration, where multiple autonomous systems interact directly with a project’s codebase. This model-level specialization—training AI to function within the real environment it will operate in—represents a significant step toward practical, autonomous software development. Composer is not trained only on text data or static code, but within a dynamic IDE that mirrors production conditions. Rush described this approach as essential to achieving real-world reliability: the model learns not just how to generate code, but how to integrate, test, and improve it in context. What It Means for Enterprise Devs and Vibe Coding With Composer, Cursor is introducing more than a fast model—it’s deploying an AI system optimized for real-world use, built to operate inside the same tools developers already rely on. The combination of reinforcement learning, mixture-of-experts design, and tight product integration gives Composer a practical edge in speed and responsiveness that sets it apart from general-purpose language models. While Cursor 2.0 provides the infrastructure for multi-agent collaboration, Composer is the core innovation that makes those workflows viable. It’s the first coding model built specifically for agentic, production-level coding—and an early glimpse of what everyday programming could look like when human developers and autonomous models share the same workspace.

The new in-house model promises 4X speed boost using reinforcement learning. It is designed for 'agentic' workflows, running tests and fixing errors autonomously.

Cursor's Composer LLM Aims to Revolutionize AI Coding

Anthropic scientists hacked Claude’s brain — and it noticed. Here’s why that’s huge

Anthropic researchers have achieved a milestone by injecting concepts like "betrayal" into Claude AI's neural networks, prompting the model to report experiencing intrusive thoughts. This experiment demonstrates the first rigorous evidence of genuine introspection in large language models, challenging existing assumptions about their capabilities. The development arrives as AI systems play an increasing role in consequential decisions, and the ability to understand AI reasoning is paramount. While this advancement could lead to more transparent and accountable AI, researchers caution that Claude's introspective abilities remain unreliable and highlight potential risks of sophisticated AI deception.

Researchers successfully injected thoughts into Claude, leading to surprising self-reporting capabilities. This breakthrough could reshape AI transparency, but also raises deception risks.

Anthropic's Claude AI Shows Signs of Self-Awareness

From static classifiers to reasoning engines: OpenAI’s new model rethinks content moderation

OpenAI has launched gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, two open-weight models under a permissive Apache 2.0 license, designed to provide enterprises with a more adaptable approach to AI safety. Unlike traditional methods that "bake in" safety policies before deployment, these models use reasoning to interpret developer-provider policies at inference time. This allows for iterative revisions and adjustments, ensuring that AI models align with evolving safety standards and business needs. While the models demonstrate strong performance, concerns remain about the centralization of safety standards and the lack of a fully open-source base model for complete iteration.

New open-source models offer enterprises greater flexibility in setting and refining safety policies, moving away from static classifiers.

OpenAI Unveils AI Model for Adaptive Content Moderation

Nvidia researchers unlock 4-bit LLM training that matches 8-bit performance

Nvidia has introduced NVFP4, an innovative technique that enables the training of large language models (LLMs) in a 4-bit quantized format without sacrificing stability or accuracy. NVFP4 not only surpasses existing 4-bit formats but also equals the performance of the larger 8-bit FP8 format, all while significantly reducing memory and computational demands. This breakthrough addresses a key challenge in model quantization, which reduces the computational and memory costs of AI models by converting parameters to lower-precision formats. By using a sophisticated multi-level scaling approach and a mixed-precision strategy, NVFP4 overcomes the limitations of previous 4-bit formats, allowing for more efficient and cost-effective training and deployment of AI models across various industries.

Nvidia researchers have developed a novel 4-bit training method, NVFP4, that matches 8-bit performance, halving memory usage and reducing compute costs. This innovation promises to democratize AI model training, making it accessible to more organizations.

Nvidia's NVFP4: 4-Bit Training Breakthrough for Large Language Models

Google adds Gemini AI app builder to AI Studio

Google has launched an AI app builder within its AI Studio, enabling users to develop AI applications using only text prompts. This positions Google alongside industry leaders like Microsoft and OpenAI, who offer similar tools for no-code app development. The new builder deeply integrates with Google's Gemini models, allowing users to incorporate image generation with the "Nano Banana" model and Gemini’s voice synthesis into their apps. Developers can visually adjust designs and guide edits with text instructions, and can plug in their own API keys after exhausting the free quota. A revamped app gallery provides templates and ideas to help users begin building immediately.

New tool allows users to create AI-powered applications with simple text prompts, putting Google in competition with Microsoft and OpenAI.

Google's AI Studio Update: App Building for Everyone

Google launches Pomelli, an AI-powered tool that creates social media campaigns automatically

Google has launched Pomelli, an AI-powered tool designed to automate the creation of social media campaigns for small and midsize businesses. By scanning a company’s website, Pomelli establishes a brand profile including tone, colors, images, and fonts, then suggests tailored campaign ideas. The generated content is editable within the interface and downloadable for immediate use, dramatically reducing the time and expense typically associated with marketing material production, according to Google. Pomelli is currently available as a public beta in the United States, Canada, Australia, and New Zealand through Google Labs, inviting users in these regions to explore its capabilities.

Pomelli, now in public beta, helps small businesses generate marketing content, reducing time and costs. The tool analyzes brand profiles to suggest campaign ideas.

Google's Pomelli: AI Tool Automates Social Media Campaigns

OpenAI targets full-scale autonomous AI researcher by early 2028

OpenAI has announced its goal to develop a fully autonomous AI researcher by early 2028 as part of a broader internal restructuring. CEO Sam Altman and Chief Scientist Jakub Pachocki shared details about the company's future infrastructure and safety strategies, emphasizing a shift toward AI tools that assist humans. OpenAI plans to develop an AI system with 'research intern' capabilities by September 2026, which will accelerate scientific research. To ensure safety, Pachocki introduced a five-layer model that includes value alignment, goal alignment, reliability, adversarial robustness, and systemic safety. The company has committed to $1.4 trillion in infrastructure spending and is collaborating with other AI labs on safety standards and transparency.

The company has revealed a detailed AI development timeline, outlining a new five-layer safety architecture and plans for massive infrastructure investments.

OpenAI Aims for Autonomous AI Researcher by 2028

Grammarly is changing its name to Superhuman

Grammarly is now Superhuman, incorporating Coda, Superhuman Mail, and the new Superhuman Go AI assistant. Available to Grammarly Pro subscribers, this suite marks a shift from writing assistance to a comprehensive AI work platform. Superhuman Go, an enhanced version of Grammarly's generative AI, connects to over 100 apps to offer contextual help, like scheduling meetings and refining pitches. This rebrand, driven by the acquisitions of Superhuman Mail and Coda, emphasizes Superhuman's focus on AI agent-powered productivity, offering assistance across various applications and web browsers.

The writing assistant evolves into a broader AI work platform, integrating with multiple apps to offer contextual help. The move reflects a shift towards AI-powered productivity across web browsers.

Grammarly Rebrands as Superhuman, Expands AI Capabilities

OpenAI restructures under new foundation, Microsoft takes 27 percent stake

OpenAI has finalized a major recapitalization, simplifying its corporate structure under the OpenAI Foundation, which now controls the OpenAI Group PBC with $130 billion in shares. This setup ensures profits from commercial activities fund its nonprofit mission, with the foundation investing $25 billion in health data and AI safety. Microsoft now holds approximately 27 percent of OpenAI, valued at $135 billion. The restructuring, influenced by attorneys general and internal tensions over safety versus growth, aims to balance commercial success with the company’s original mission of benefiting humanity, particularly after CEO Sam Altman's brief ouster and key researcher departures.

The AI company completes a major restructuring, with its nonprofit arm now controlling a Public Benefit Corporation. Microsoft secures a significant minority stake, redefining their financial partnership.

OpenAI Restructures: Foundation Takes Control, Microsoft Gets Stake

Microsoft’s Copilot can now build apps and automate your job — here’s how it works

Microsoft is significantly expanding its Copilot AI assistant, introducing tools that empower employees to build applications, automate workflows, and create specialized AI agents using conversational prompts. The new 'App Builder' and 'Workflows' features allow 100 million Microsoft 365 users to develop business tools as easily as drafting emails, marking a major step in merging AI with software development. Users can now describe an application, and Copilot will generate a working app with a database, UI, and security controls. This initiative is part of Microsoft's broader strategy to integrate AI across its product portfolio, driven by its partnership with OpenAI, and could potentially turn millions of office workers into part-time software developers.

New tools enable users to create applications and automate tasks using natural language, potentially democratizing software development.

Microsoft's Copilot Aims to Transform Workers into App Builders

GitHub's Agent HQ aims to solve enterprises' biggest AI coding problem: Too many agents, no central control

GitHub is tackling the challenge of managing multiple AI coding agents with its new Agent HQ. Announced at Universe 2025, this architecture transforms GitHub into a unified control plane, enabling enterprises to manage AI agents from various vendors like Anthropic, OpenAI, and Google. Instead of forcing developers to use a single agent, GitHub aims to be the essential layer that orchestrates them all. Agent HQ allows developers to work with Git, pull requests, and issues while using agents from multiple vendors within GitHub's security perimeter, ensuring enterprise-grade governance. The platform introduces Mission Control, a unified command center accessible across GitHub's interfaces, which allows developers to assign tasks to multiple agents and manage permissions from a single point. Custom agents, defined via AGENTS.md files, allow enterprises to codify organizational standards, ensuring consistent output quality across teams, while native Model Context Protocol (MCP) support integrates external services without requiring individual agent integrations.

New architecture unifies multiple AI coding agents, offering a single control plane for improved security and collaboration within enterprise environments.

RecursivAI

Latest News:

Anthropic's Claude AI Shows Signs of Self-Awareness

AI Revolutionizes Mathematics: New Initiative Launches

Cursor's Composer LLM Aims to Revolutionize AI Coding

AI 101: Zero-shot & Few-shot Learning

Rexommendation

Limited Time Offer

FREE LIFETIME SUBSCRIPTION

Join Rex's Growing Pack!