Top AI Companies Agree to Work with Pentagon on Secret Data

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 👇

The IIoT Postgres Limits No One Talks About Until Production

Most IIoT teams don't realize Postgres is at its limit until queries start failing in production.

Our new white paper, The IIoT PostgreSQL Performance Envelope, maps exactly where Postgres hits its limits with industrial sensor data and what you can do before you're forced into a split architecture. No hand-waving. Real benchmarks, real query patterns, real thresholds.

If you're building on IIoT telemetry and still deciding whether Postgres can scale with you, this is the data you need.

Access the data

print("News & Trends")

Top AI Companies Agree to Work with Pentagon on Secret Data (4 min. read)

The Pentagon has inked agreements with seven leading AI firms, including Microsoft, Amazon, and Google, to integrate their technologies into classified military networks. This move aims to enhance battlefield decision-making through advanced data analysis. Notably, Anthropic, a key industry player, remains sidelined after disputes over the use of its AI for surveillance and autonomous weapons. While some employees express ethical concerns, these partnerships signify a deepening collaboration between tech giants and the defense sector.

Anthropic Tests Jupiter V1 Ahead of Developer Conference (2 min. read)

Anthropic is internally testing a new AI model, codenamed "Jupiter V1," ahead of its "Code with Claude" conference on May 6. This follows their pattern of using planetary names for pre-release models, as seen with "Neptune" before the Claude 4 launch. The timing suggests a significant announcement is imminent, possibly introducing a new model tier or enhancements to existing ones. The choice of "Jupiter" hints at a substantial upgrade, aligning with Anthropic's commitment to rigorous safety evaluations before deployment.

Anthropic’s Co-founder Predicts AI Self-building Era (5 min. read)

Jack Clark explores the imminent reality of AI systems autonomously conducting research and development, potentially leading to self-improvement cycles by 2028. He highlights advancements like Claude Mythos Preview's 93.9% success rate on SWE-Bench, underscoring AI's growing proficiency in complex coding tasks. Clark emphasizes that while full automation isn't expected in 2026, the trajectory suggests a transformative shift in AI's role in innovation, urging the community to prepare for the profound implications of this evolution.

print("Applications & Insights")

Optimizing ML Workload Network Efficiency (Part I): Feature Trimmer (14 min. read)
Pinterest's ML serving system, structured in a root-leaf architecture, faced network bottlenecks due to excessive feature transmission between components. To address this, they implemented Feature Trimmer, a solution that compresses and prunes redundant features, significantly reducing network load. This optimization not only alleviated bandwidth constraints but also enhanced GPU utilization, leading to improved system efficiency and cost savings.

How We Rebuilt Search Ranking at Faire with Deep Learning (5 min. read)
Faire's engineering team overhauled their search ranking system by integrating deep learning, transitioning from traditional methods to a more sophisticated, data-driven approach. This shift led to significant improvements in search relevance and user engagement, demonstrating the power of deep learning in enhancing e-commerce search experiences.

How We Automated Data Validation (5 min. read)
Hex's data team revolutionized their validation process by developing an AI-powered audit agent, enhancing speed, consistency, and depth. This agent automates standard checks and delves into complex root cause analyses, all while allowing human oversight for nuanced cases. The result is a seamless, customizable validation workflow that empowers data professionals to focus on impactful work, ensuring robust and reliable data artifacts across the organization.

Handling Schema Issues in Polars (5 min. read)
Schema changes can disrupt data pipelines, manifesting as new columns, missing columns, type drift, or breaking changes like renaming. Polars offers tailored solutions for each scenario across formats like CSV, Parquet, Delta Lake, and Apache Iceberg. For instance, in CSVs, you can use schema_overrides to specify column types, while in Parquet, parameters like missing_columns="insert" handle schema evolution. Understanding these tools ensures robust and adaptable data processing.

print("Tools & Resources")

TRENDING MODELS

Text Generation
deepseek-ai/DeepSeek-V4-Pro
⇧ 631k Downloads
DeepSeek-V4-Pro is a 862-billion parameter model designed for advanced text generation tasks. It offers state-of-the-art performance in generating coherent and contextually relevant text across various applications.

Token Classification
openai/privacy-filter
⇧ 141k Downloads
The privacy-filter model by OpenAI is a 1-billion parameter model aimed at identifying and filtering sensitive information in text. It enhances data privacy by detecting and masking personal or confidential data in textual content.

Text Generation
mistralai/Mistral-Medium-3.5-128B
⇧ 15k Downloads
Mistral-Medium-3.5-128B is a 128-billion parameter model developed for high-quality text generation. It excels in producing fluent and contextually appropriate text for a wide range of applications.

Text Generation
XiaomiMiMo/MiMo-V2.5-Pro
⇧ 13.3k Downloads
MiMo-V2.5-Pro is a 1-trillion parameter model from XiaomiMiMo, designed for large-scale text generation tasks. It delivers exceptional performance in generating diverse and contextually accurate text outputs.

Any-to-Any
nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16
⇧ 44.6k Downloads
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 is a 33-billion parameter model by NVIDIA, capable of handling various tasks with advanced reasoning capabilities. It supports multiple applications, including text generation and comprehension tasks.

TRENDING AI TOOLS

• 📊 Datanomy: A Python package for data profiling and exploration.

• 🌐 Palma: The enterprise layer that unlocks MCP at scale

• 🚀 Cofounder 2: Run an entire company with agents.

print("Everything else")

• Google's Gemini AI now outputs answers directly as documents, PDFs, and Excel files.

• YouTube is testing an AI-powered search feature to provide guided answers.

• Anthropic is in talks to acquire AI chips from a UK startup to enhance its computational capabilities.

• Panthalassa secures $140 million funding to enhance AI capabilities for maritime applications.

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