2025 is already 25% over, and you’re not even 10% closer to your goals.
But here’s your moment to flip the script.

It’s time to learn the most in-demand skill of 2025 - AI -and finally take control of your time, growth, and impact.

Save your free spot here (only 100 free seats)

Join this FREE 3-hour AI Training (worth $399) - designed to help you master 20+ powerful AI tools and prompting techniques that can save you up to 16 hours a week.

🚀 But there’s a catch: Only the first 100 people get it free.

This hands-on course will teach you how to:

👉 Automate tasks and save hours – Streamline your workflow and focus on what truly matters.
👉 Make smarter, faster decisions – Use AI-driven insights to power up your business or career.
👉 Grow your personal brand with AI – Create high-impact LinkedIn content in minutes.
👉 Write like a pro with ChatGPT – Emails, reports, presentations… done in a fraction of the time.

⏳ Spots are limited.
🎁 First 100 seats are absolutely FREE.

Ready to level up?
Grab your free spot now before it's gone.

Sponsored

Subscribe | Submit a tip | Advertise with us

Welcome to DataPro 133 – Top Tools/Datasets Driving New Research 🔧📊, your weekly download on the breakthroughs redefining what AI and data teams can do. From OpenThoughts2-1M and Llama-Nemotron to DeepSeek-V3 and Meta’s Maverick, these new releases aren’t just datasets, they’re accelerators for reasoning, coding, and multimodal exploration.

⚡ Also making waves: a bold new upgrade for data pros who’ve hit the Tableau plateau. The Tableau Cookbook for Experienced Professionals is now available for pre-order offering performance tuning, enterprise-ready governance, and the interactive magic your dashboards deserve.

Whether you're scaling models, building agents, or sharpening your BI stack, this edition is stacked with what's next. Let’s dive in.

Cheers,

Merlyn Shelley

Growth Lead, Packt

Tableau Cookbook for Experienced Professionals
Now available for pre-order | Shipping April 25, 2025

The Tableau Plateau: Why So Many Get Stuck

At first, Tableau feels like magic.
Drag, drop, and suddenly, your data tells a story.

But fast forward a year, and the sparkle starts to fade:

- Dashboards are slow and clunky

- Your filters conflict, your data models sprawl

- Stakeholders ask for secure access, and you realize you’ve hit a wall

It’s not that Tableau can’t do it.
It’s that you’ve outgrown the basics.

🔧 What Got You Here Won’t Get You There

That’s where Tableau Cookbook for Experienced Professionals steps in. Written by two experts who have trained Fortune 500 teams, led global analytics initiatives, and built enterprise-scale BI systems, this book offers a real-world-tested path to next-level Tableau mastery.

👤 Pablo Sáenz de Tejada – Snowflake, Salesforce, The Information Lab
👤 Daria Kirilenko – DSCOVR Analytics, Stanford University

They’ve seen it all - from confident dashboard dabblers to elite data professionals. And they know the steps it takes to bridge that gap.

🚀 The Three Shifts Every Advanced User Must Make

Performance

Go beyond visual appeal - build dashboards that are lightning fast and designed for scale.
Learn:

- Data model optimization

- Tableau Cloud’s Data Management features

- Performance troubleshooting with built-in tools

Interactivity

Stop creating dashboards that “look good.” Start building tools users love to explore.
Learn:

- Zone visibility and advanced UX workflows

- LOD expressions and table calculations

- Layered interactivity through dynamic filters and tooltips

Governance

Master Tableau in the enterprise arena. Secure it. Scale it. Own it.
Learn:

- REST API and TabPy integrations

- Enterprise security strategies

- Tableau’s Content Migration Tool (2025.1 and beyond)

🛠️ Real-World Impact in Action

A global retailer’s dashboards were bloated and untrustworthy. After applying this book’s spatial join techniques and content structuring strategies, they reduced load time by 50%, streamlined permissions, and uncovered regional gaps in real-time sales.

This book isn’t about “more charts.”
It’s about building tools that drive real business decisions.

✅ What You’ll Unlock

Hands-on recipes (60+) from senior consultants

- Frameworks for troubleshooting, performance, and secure deployment

- Advanced topics like TabPy, APIs, and scalable data modeling

- A PDF eBook with purchase for on-the-go access

🔓 Ready to Break Through?

📅 Release Date: April 25, 2025
🎁 Bonus templates and code samples for early buyers
💡 Free PDF eBook with Kindle or print purchase

⭕ deepseek-ai/DeepSeek-V3-0324: DeepSeek introduced V3-0324 with enhanced reasoning (MMLU-Pro +5.3, GPQA +9.3, AIME +19.8), better code execution, improved Chinese writing, refined translation, more accurate function calling, and detailed search analysis. New system prompt and optimized temperature mapping included.

⭕ ByteDance/InfiniteYou: ByteDance introduced InfiniteYou (InfU), leveraging Diffusion Transformers (DiTs) like FLUX for high-fidelity, identity-preserved image generation. InfU improves identity similarity, text-image alignment, and aesthetics using InfuseNet and multi-stage training. Two model variants, aes_stage2 (better aesthetics) and sim_stage1 (higher ID similarity), enhance flexibility.

⭕ manycore-research/SpatialLM-Llama-1B: SpatialLM introduced SpatialLM-Llama-1B, a 3D large language model that processes point cloud data to generate structured 3D scene understanding. It identifies architectural elements (walls, doors, windows) and object bounding boxes. It supports multimodal inputs, enhancing applications in robotics and navigation.

⭕ canopylabs/orpheus-3b-0.1-ft: Canopy Labs introduced Orpheus 3B 0.1 FT, a Llama-based speech model fine-tuned for high-quality, empathetic text-to-speech generation. It offers human-like intonation, zero-shot voice cloning, guided emotions, and low-latency real-time streaming, making it ideal for natural speech synthesis applications.

⭕19 Git Tips For Everyday Use: The post shares practical Git commands and techniques to improve workflow efficiency. It covers logging, file extraction, rebasing, managing branches, fixing commits, using aliases, and troubleshooting, offering valuable insights for intermediate Git users.

⭕ AI Expert Roadmap: This post offers an interactive collection of roadmaps covering AI, data science, machine learning, deep learning, and big data engineering. It guides learners on essential concepts, tools, and techniques while encouraging ongoing exploration of evolving technologies and best practices.

⭕ Cookiecutter Data Science: The Cookiecutter Data Science v2 introduces an improved, standardized project structure for data science workflows. It offers a command-line tool (ccds) that simplifies project setup and enforces best practices. With enhanced functionality and flexible directory organization, it ensures consistency and reproducibility across projects.

⭕ Google DeepMind Researchers Propose CaMeL: A Robust Defense that Creates a Protective System Layer around the LLM, Securing It even when Underlying Models may be Susceptible to Attacks. Google DeepMind introduces CaMeL, a security layer that protects LLMs from prompt injection attacks without modifying the underlying models. Using a dual-model architecture and metadata-based policies, CaMeL isolates untrusted data, ensuring safer decision-making and outperforming existing defenses in security and reliability.

⭕ A Code Implementation for Advanced Human Pose Estimation Using MediaPipe, OpenCV and Matplotlib: This tutorial demonstrates advanced human pose estimation using MediaPipe, OpenCV, and Matplotlib. It guides developers through detecting, visualizing, and extracting keypoints from images, enabling applications in sports, healthcare, and interactive systems. The code efficiently processes and annotates pose landmarks with high accuracy.

⭕ Sea AI Lab Researchers Introduce Dr. GRPO: A Bias-Free Reinforcement Learning Method that Enhances Math Reasoning Accuracy in Large Language Models Without Inflating Responses: Sea AI Lab introduces Dr. GRPO, a bias-free reinforcement learning method that improves LLMs’ math reasoning accuracy without inflating responses. It eliminates response-length biases, ensuring fair model updates. Dr. GRPO-trained models outperformed others on key benchmarks while maintaining efficiency and reducing unnecessary verbosity.

⭕ Anyscale powers AI compute for any workload using Google Compute Engine: Anyscale, built on Google Compute Engine (GCE) and Kubernetes Engine (GKE), powers scalable AI workloads across diverse environments. By optimizing compute flexibility and performance, it enables efficient model training, inference, and deployment. Anyscale reduces costs, boosts GPU utilization, and ensures reliable AI scaling across industries.

⭕ Formula E’s AI equation: A new Driver Agent for the next generation of racers. Formula E partners with Google Cloud to introduce the AI-powered Driver Agent, leveraging Vertex AI and Gemini to analyze multimodal racing data. This tool democratizes access to data-led coaching, helping aspiring drivers refine performance by comparing their laps with professional benchmarks.

⭕ Nuro drives autonomous innovation with AlloyDB for PostgreSQL: Nuro enhances autonomous vehicle innovation by migrating to AlloyDB for PostgreSQL, enabling seamless data management, high query performance, and vector similarity searches. This transition reduces operational costs, accelerates AI model training, and ensures continuous improvement of autonomous driving systems across complex real-world scenarios.

⭕ Enhance deployment guardrails with inference component rolling updates for Amazon SageMaker AI inference: Amazon SageMaker AI introduces rolling updates for inference components, enhancing model deployment by reducing resource overhead, preventing downtime, and enabling batch-based updates with automatic rollback safeguards. This feature optimizes resource use and ensures reliable, cost-effective updates for GPU-heavy workloads, maintaining high availability in production environments.

⭕ Integrate natural language processing and generative AI with relational databases: Amazon introduces a solution integrating natural language processing (NLP) and generative AI using Amazon Bedrock and Aurora PostgreSQL. It enables users to query relational databases using conversational language, reducing SQL complexity, democratizing data access, and easing the burden on developers through AI-driven SQL generation.

⭕ Automate Supply Chain Analytics Workflows with AI Agents usingn8n: n8n revolutionizes supply chain analytics by enabling AI-powered workflow automation without extensive coding. Using pre-built nodes, users can build AI agents to process emails, generate SQL queries, and update databases. This low-code platform empowers non-technical teams to maintain and enhance workflows efficiently.

⭕ Uncertainty Quantification in Machine Learning with an Easy Python Interface: ML Uncertainty is a Python package that simplifies uncertainty quantification (UQ) for machine learning models, providing reliable prediction intervals with minimal code. Built on top of SciPy and scikit-learn, it enables users to estimate uncertainties efficiently, enhancing model interpretability and real-world decision-making.

⭕ The Ultimate AI/ML Roadmap for Beginners: This post guides aspiring professionals through the essential steps to master AI and machine learning. Covering math fundamentals, Python, data structures, and algorithms, this roadmap equips learners to apply AI/ML in real-world scenarios without requiring a PhD.

⭕ Attractors in Neural Network Circuits:Beauty and Chaos. This article explores how neural networks, when modeled as dynamical systems, evolve over time and converge to attractors, fixed points, limit cycles, or chaotic patterns. By adding feedback loops and nonlinear activations, even simple neural networks generate intricate behaviors, offering insights into memory formation, oscillating reactions, and chaotic processes.

⭕ Least Squares: Where Convenience Meets Optimality. Least Squares is the cornerstone of regression models, primarily because of its simplicity, mathematical optimality, and deep connection with Maximum Likelihood Estimation (MLE). Beyond its computational ease, it minimizes Mean Squared Error (MSE) efficiently, derives the mean as a natural consequence of L2 minimization, and provides the Best Linear Unbiased Estimator (BLUE) when applied to Ordinary Least-Squares (OLS).