GitHub Open Sourced Their MCP Server

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...

But first, a quick message from our partner 👇

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

print("News & Trends")

Why GitHub Open Sourced Their MCP Server and What It Means for You (5 min. read)

GitHub has open sourced its Model Context Protocol (MCP) server, a standardized interface that lets LLMs interact smoothly with external tools and data sources. This move enables applications like Copilot to securely access codebases, APIs, and enterprise systems, fostering interoperability and trust. By opening MCP, GitHub invites the community to build richer AI ecosystems where models can reason, act, and integrate more effectively across diverse environments.

Microsoft Brings AI to Your Formulas with the Copilot Function in Excel (5 min. read)

Microsoft's Copilot in Excel now offers an "Explain Formula" feature, providing step-by-step breakdowns of complex formulas directly within your spreadsheet. This AI-driven tool delivers contextual explanations grounded in your actual data, enhancing comprehension without requiring external documentation. By turning opaque formulas into clear narratives, it helps users learn, debug, and trust their calculations, bridging the gap between advanced functionality and everyday usability.

Deep Think with Confidence (2 min. read)

Deep Think with Confidence (DeepConf) introduces a parallel thinking approach that strengthens LLM reasoning by filtering out low-confidence traces during or after generation using internal confidence signals. The method requires no retraining or hyperparameter tuning and integrates directly into existing frameworks. Tested on the AIME 2025 dataset, DeepConf achieved up to 99.9% accuracy, showcasing a lightweight yet powerful way to boost reasoning reliability.

Elon Musk Open Sources Grok 2.5 and Announces Plans for Grok 3 (1 min. read)

xAI has open sourced the Grok 2.5 model on Hugging Face, giving researchers and developers full access to its weights and architecture. Elon Musk also revealed that Grok 3 is on track for open source release in about six months, signaling a rapid iteration cycle and continued push to compete in the frontier AI space.

print("Applications & Insights")

Advanced Prompt Engineering for Data Science Projects (11 min. read)
This article delves into advanced prompt engineering strategies to boost LLM performance in data science workflows. It explores few-shot learning to improve comprehension, chain-of-thought prompting for stepwise reasoning, and role prompting to align model behavior with domain expertise. The piece also highlights prompt chaining and evaluation techniques, offering practitioners practical ways to refine outputs and tackle complex analytical challenges with greater reliability.

Enhancing Model Safety through Pretraining Data Filtering (5 min. read)
Anthropic's latest research demonstrates that filtering harmful content from pretraining datasets can significantly reduce a language model's ability to generate dangerous information, such as details on CBRN weapons, without compromising its overall performance. By employing classifiers to identify and exclude harmful documents, models pretrained on this curated data showed a 33% decrease in harmful capabilities evaluations, while maintaining proficiency in standard benchmarks like MMLU and coding tasks.

Build vs Buy in the Age of AI (5 min. read)
Marty Cagan unpacks the evolving build-versus-buy dilemma in the AI era, explaining that while generative-AI tools like Lovable and Bolt let non-technical users build apps via natural language, complex enterprise domains—think compliance, security, financial workflows—still demand robust, rule-laden systems. His conclusion: the future is hybrid—businesses will buy core components exposed via machine-readable protocols like MCP, while AI-generated workflows complement them, blending build and buy into a cohesive strategy.

How to Fix Your Context (7 min. read)
Drew Breunig tackles long-context failures in LLMs, like poisoning, distraction, confusion, and clash, and delivers six smart fixes: RAG for injecting just-right info, tool loadouts to avoid overwhelm, context quarantine via isolated threads, pruning irrelevant history, summarizing bulk context, and offloading notes externally. These techniques help agents stay sharp by managing information rigorously, because even massive context windows are no excuse for sloppy context engineering.

print("Tools & Resources")

TRENDING MODELS

Text Generation
deepseek-ai/DeepSeek-V3.1-Base
⇧ 15.3k Downloads
DeepSeek-V3.1-Base is a 685-billion parameter model designed for advanced text generation tasks.

Text Generation
ByteDance-Seed/Seed-OSS-36B-Instruct
⇧ 6.36k Downloads
Seed-OSS is a series of open-source large language models developed by ByteDance's Seed Team, designed for powerful long-context, reasoning, agent and general capabilities, and versatile developer-friendly feature

Text Generation
google/gemma-3-270m
⇧ 74.1k Downloads
Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions.

Text Generation
nvidia/NVIDIA-Nemotron-Nano-9B-v2
⇧ 28.8k Downloads
NVIDIA-Nemotron-Nano-9B-v2 is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks.

TRENDING AI TOOLS

• 📊 Julius: AI data analyst to chat with and visualize your data

• 💻 Qoder: AI-powered coding platform for real software

• ⚡ Action Agent: Enterprise AI agent for complex tasks

• 🧠 Command A Reasoning: Cohere’s advanced model for enterprise reasoning tasks.

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