Harvard’s 16TB Public Dataset

Hey there 👋

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

Here is your 5-minute pulse...

print("News & Trends")

Harvard’s Library Innovation Lab Releases 16TB Archive of 2024-2025 Federal Datasets for Public & Academic Use (2 min. read)

Harvard’s Library Innovation Lab just launched the Data.gov Archive, a 16TB collection of 311,000+ federal datasets, updated daily. This project ensures public data remains accessible and authenticated for research, policymaking, and public use. Open-source tools are included to help others replicate this effort.

AI’s Exponential Leap: Sam Altman’s Take on the Future of Intelligence (4 min. read)

In his latest blog post, Sam Altman observes that AI intelligence scales logarithmically with resources, usage costs are dropping tenfold annually, and the socioeconomic value of AI is growing super-exponentially. He predicts AI agents will soon function as virtual co-workers, significantly transforming society and the economy.

DeepMind claims its AI performs better than International Mathematical Olympiad gold medalists (6 min. read)

Google DeepMind’s AI, AlphaGeometry2, has outperformed the average gold medalist in solving International Mathematical Olympiad geometry problems. By blending neural networks with symbolic reasoning, it solved 84% of past problems, hinting at AI’s future in advanced problem-solving. Could this hybrid approach be the key to more capable general AI?

Andrew Ng introduces Agentic Object Detection

Andrew Ng introduces a reasoning-driven approach to object detection, enabling text-prompt-based detection without labeling or training. By leveraging advanced reasoning, it identifies complex objects and scenarios, making rapid prototyping and deployment seamless. While processing takes 20–30 seconds per image, ongoing improvements aim to enhance speed and efficiency.

print("Applications & Insights")

10 Useful LangChain Components for Your Next RAG System (4 min. read)
Learn 10 key LangChain components for building effective Retrieval-Augmented Generation (RAG) systems. You'll go through how to use Document Loaders, Text Splitters, Embeddings, Vector Stores, and more to enhance LLM-powered applications with better data retrieval and contextual accuracy.

Measuring LLM similarity (6 min. read)
A new study introduces Chance Adjusted Probabilistic Agreement (CAPA) to measure error similarity across language models. Findings show that as models improve, their mistakes become more alike—raising concerns for AI oversight and diversity. Understanding these patterns is key to building more robust AI systems.

Synthetic Data Generation with LLMs (9 min. read)
This tutorial walks through using LLMs to generate synthetic data for evaluating Retrieval-Augmented Generation (RAG) systems. It covers prompting techniques, dataset structuring, and automation strategies to create high-quality test data, reducing manual effort while improving AI evaluation and scalability.

print("Tools & Resources")

TRENDING MODELS

Reinforcement Learning
ValueFX9507/Tifa-Deepsex-14b-CoT-GGUF-Q4
⇧ 40.3k Downloads
This model integrates chain-of-thought reasoning into reinforcement learning frameworks, enhancing decision-making processes in complex environments.

Text-to-Speech
hexgrad/Kokoro-82M
⇧ 300k Downloads
Kokoro-82M is a compact text-to-speech model that delivers natural and expressive voice synthesis, suitable for applications with limited computational resources.

Text Generation
simplescaling/s1-32B
⇧ 5.39k Downloads
s1-32B is a large-scale language model designed for high-quality text generation, leveraging extensive training data to produce accurate and contextually relevant outputs.

Text-to-Image
black-forest-labs/FLUX.1-dev
⇧ 1.58M Downloads
FLUX.1-dev is an advanced text-to-image model that generates high-quality visuals from textual descriptions, useful for creative and design applications.

Text Generation
mistralai/Mistral-Small-24B-Instruct-2501
⇧ 263k Downloads
Mistral-Small-24B-Instruct-2501 is fine-tuned for instructional text generation, making it adept at creating educational content and step-by-step guides.

Note: Multiple DeepSeek models such as the DeepSeek-R1 and Janus-Pro-7B are still trending. Not shown here to highlight other useful models.

TRENDING AI TOOLS

• 🤖 Github Copilot: Microsoft’s coding assistant with new agentic features

• 💻 Warp: AI in your terminal, providing guidance on the next command to run.

• 🗣️ Speak: Transcription, translation, and analysis for audio, video, and text.

• 🛠️ Taylor:Automate and enrich freeform text for business and engineering.

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