Anthropic Integrates Interactive MCP Apps into Claude

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science this week.

Here is your 5-minute pulse...

print("News & Trends")

Anthropic Integrates Interactive MCP Apps into Claude (1 min. read)

Anthropic has supercharged Claude by embedding interactive productivity tools like Asana, Slack, Figma, and Box directly into its interface. Now, users can manage projects, draft messages, design diagrams, and handle files without ever leaving Claude. This leap forward utilizes the Model Context Protocol (MCP) and its extension, MCP Apps, enabling real-time collaboration within the chat environment. By open-sourcing MCP, Anthropic invites developers to craft robust integrations, fostering a dynamic ecosystem for AI-driven workflows.

Prep for the SAT with Practice Tests in Gemini (2 min. read)

Google's Gemini app now offers free, full-length SAT practice tests, thanks to a partnership with The Princeton Review. These tests provide immediate feedback, pinpointing your strengths and areas for improvement. Gemini can also explain answers and help craft personalized study plans, making your SAT prep more efficient and tailored to your needs.

GitHub Copilot SDK (2 min. read)

GitHub's Copilot SDK empowers developers to seamlessly integrate the Copilot Agent into their applications and services across multiple platforms. By automating tasks like request verification, response formatting, and API interactions, the SDK allows you to focus on building core functionalities without the hassle of boilerplate code. Whether you're working in Node.js, Python, or .NET, this toolkit streamlines the development of AI-powered features, making your apps smarter and more responsive.

Pandas 3.0 Released! (3 min. read)

Pandas 3.0 is here, bringing a slew of enhancements that data scientists and ML engineers will appreciate. String columns now default to the str dtype, boosting performance and type safety. The introduction of Copy-on-Write (CoW) eliminates the notorious SettingWithCopyWarning, ensuring more predictable behavior and better performance by avoiding unnecessary copies. Additionally, datetime-like data now defaults to microsecond resolution, reducing out-of-bounds errors. There's also a new pd.col() syntax for streamlined DataFrame.assign operations. Be aware, these changes might require updates to your existing code. For a comprehensive list of improvements and guidance on upgrading, check out the release notes.

Claude in Excel (2 min. read)

Claude in Excel, now in beta for Pro, Max, Team, and Enterprise users, seamlessly integrates AI into your spreadsheet workflow. It offers cell-level insights, preserves formula dependencies during assumption updates, and swiftly debugs errors. With support for pivot tables, charts, and file uploads, Claude enhances efficiency in financial modeling and complex data analysis.

print("Applications & Insights")

Small Models, Big Results: Achieving Superior Intent Extraction Through Decomposition (5 min. read)
Google researchers introduce a two-stage method for understanding user intents from UI interactions using small, on-device models. First, each screen interaction is individually summarized. Then, these summaries are aggregated to extract the overall user intent. This decomposition allows smaller models to achieve performance comparable to larger ones, enabling efficient and privacy-preserving intent recognition directly on mobile devices.

Salesforce Ships Higher-Quality Code Across 20,000 Developers with Cursor (3 min. read)
Salesforce has integrated Cursor into its development workflow, with over 90% of its 20,000 engineers now using the tool daily. This widespread adoption has led to double-digit improvements in cycle time, pull request velocity, and code quality. Initially embraced by junior engineers to better understand existing codebases, senior engineers soon recognized Cursor's value in automating tedious tasks, leading to broader use across teams. The result is a significant enhancement in product quality and a more efficient development process.

How to Cut RAG Latency in Half
This executive playbook delves into the architectural tweaks and model strategies that slashed real-world Retrieval-Augmented Generation (RAG) latency by over 50%. It offers practical insights on identifying major latency culprits, such as the effects of query rewriting, and explores how parallel model execution can enhance both speed and user experience. By comparing before-and-after architectures and analyzing real benchmark data, the guide provides actionable takeaways for engineering teams aiming to boost their RAG systems' responsiveness without compromising quality.

How I Reduced AI Token Costs by 91% with Semantic Tool Selection and Redis (5 min. read)
Facing soaring LLM costs from sending all 70 tool definitions per query, the author implemented a semantic tool selection system using Redis as a vector database. By creating multi-component embeddings that capture tool descriptions, parameters, usage examples, and return types, the system intelligently matches user queries to relevant tools. This approach slashed token consumption by 91%, reduced costs by 49%, and improved accuracy, leading to faster responses and enhanced user satisfaction.

Path Forward for Data Governance: Existence Over Essence (8 min. read)
Traditional data governance, rooted in the belief that data holds inherent meaning, falls short in AI-driven environments where context dictates value. Winfried Etzel argues for a shift from controlling data to intentionally structuring socio-technical systems, ensuring sustained meaning and accountability. This existential reframing is essential for governing AI at scale.

print("Tools & Resources")

TRENDING MODELS

text-to-image
zai-org/GLM-Image
⇧ 13.9k Downloads
GLM-Image is a text-to-image model developed by zai-org, capable of generating high-quality images from textual descriptions. It leverages advanced generative techniques to produce detailed and contextually relevant visuals.

text-generation
zai-org/GLM-4.7
⇧ 65.2k Downloads
GLM-4.7 is a large-scale text generation model by zai-org, designed to produce coherent and contextually accurate text across various domains. With 358 billion parameters, it excels in generating human-like responses for diverse applications.

image-to-video
Lightricks/LTX-2
⇧ 2.38M Downloads
LTX-2 by Lightricks is an image-to-video model that transforms static images into dynamic video sequences. It utilizes advanced algorithms to create smooth and realistic animations from single-frame inputs.

text-to-image
Tongyi-MAI/Z-Image-Turbo
⇧ 538k Downloads
Z-Image-Turbo, developed by Tongyi-MAI, is a text-to-image model that generates high-resolution images from textual prompts. It focuses on speed and quality, delivering detailed visuals efficiently.

text-generation
Alibaba-Apsara/DASD-4B-Thinking
⇧ 2.09k Downloads
DASD-4B-Thinking by Alibaba-Apsara is a 4-billion parameter text generation model designed for thoughtful and context-aware text production. It aims to provide nuanced and accurate responses in various conversational scenarios.

TRENDING AI TOOLS

• 📊 CharGen: Create stunning charts effortlessly with AI-powered design and customization.

• 📝 pynb: Minimalist notebook interface for Python with a focus on simplicity and efficiency.

• 🤖 Slackbot: Context-aware AI agent enhancing productivity in Slack."

print("Everything else")

• This article shows how to turn todos into tasks in Claude Code.

• Google introduces Personal Intelligence AI mode in Search to enhance personalized query results and recommendations.

• Learn how to optimize data transfer in distributed AI/ML training workloads.

That’s it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve the pulse experience for you.

See you soon,

Andres