Google to Invest Up to $40 Billion in Anthropic

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…

You think 4x faster than you type. Your IDE should keep up.

Wispr Flow lets you dictate prompts, acceptance criteria, and bug reproductions inside Cursor or Warp — with automatic file name and variable recognition. Say user_id, get user_id. Say useEffect, get useEffect.

Paste directly into GitHub, Jira, or Linear. Give coding agents the full context they need without typing a novel.

89% of messages sent with zero edits. Millions of developers use Flow daily, including teams at OpenAI, Vercel, and Clay. Free on Mac, Windows, and iPhone.

Start flowing free

print("News & Trends")

Google to Invest Up to $40 Billion in Anthropic (3 min. read)

Google is set to invest at least $10 billion in Anthropic, with the potential to increase this to $40 billion if certain performance targets are met. This move follows Amazon's recent $5 billion investment in the AI startup. Both tech giants are providing cloud computing resources and AI chips to help Anthropic scale its Claude models, which have seen a surge in demand. These investments highlight the growing trend of major companies funding AI startups to integrate their technologies and services.

DeepSeek V4 Preview Release (3 min. read)

DeepSeek has unveiled V4, featuring the Pro model with 1.6 trillion parameters and the Flash model with 284 billion, both supporting a 1 million token context length. V4-Pro matches top closed-source models in performance, while V4-Flash offers a cost-effective alternative. Innovations like token-wise compression and DeepSeek Sparse Attention enhance efficiency. Integrated with leading AI agents, DeepSeek-V4 is now accessible via chat.deepseek.com and updated APIs.

Anthropic Launches Memory in Claude Agents for Enterprise (1 min. read)

Anthropic has rolled out a public beta of the Memory feature for Claude Managed Agents, enabling these AI agents to retain and utilize information from previous sessions. This advancement allows for the accumulation of knowledge over time without manual prompt updates. Designed with a filesystem-based structure, Memory ensures data can be exported, managed via APIs, and controlled with specific permissions. Early adopters like Netflix and Rakuten are already leveraging this feature to streamline workflows and reduce errors.

print("Applications & Insights")

Background Coding Agents: Supercharging Downstream Consumer Dataset Migrations (Honk, Part 4) (6 min. read)
Spotify's engineering team faced the daunting task of migrating approximately 1,800 data pipelines to new dataset versions within six months. By leveraging their background coding agent, Honk, alongside Backstage and Fleet Management tools, they automated significant portions of this process, saving an estimated 10 engineering weeks. This case study highlights the importance of context engineering and the challenges posed by varying pipeline frameworks, emphasizing the need for standardization to enhance automation efficiency.

Measurement Engineering: The Part of Data Science That Will Thrive in AI (4 min. read)
In this insightful piece, Eric Weber highlights the growing importance of measurement engineering in data science, especially as AI automates more execution tasks. He argues that while AI excels at tasks like writing queries and building models, it falls short in areas requiring human judgment, such as determining if we're measuring the right things or interpreting ambiguous results. Weber emphasizes that the ability to discern meaningful metrics and make informed decisions is becoming the defining skill in our field. He introduces the concept of "measurement engineers"—professionals who focus on ensuring that data measurements accurately reflect the outcomes businesses care about, a role that is increasingly vital as AI handles more of the technical execution.

Vector Databases Are Dying. Here’s the Production Evidence (9 min. read)
The article examines the declining viability of standalone vector databases like Pinecone, citing escalating costs and operational complexities in production environments. It highlights cases where teams faced unexpected expenses and maintenance challenges, leading many to migrate to integrated solutions such as pgvector, which offer comparable performance at a fraction of the cost. The piece underscores a broader industry trend of embedding vector search capabilities directly into existing databases, questioning the necessity of dedicated vector databases for most applications.

Mastering Claude Code in 30 Minutes (Video)
This video by Anthropic offers a comprehensive guide to mastering Claude Code, providing practical insights and techniques to enhance your coding skills. It covers key concepts and best practices, making it a valuable resource for data scientists and ML engineers aiming to deepen their understanding of this programming language.

print("Tools & Resources")

TRENDING MODELS

Text Generation
deepseek-ai/DeepSeek-V4-Pro
⇧ 174k Downloads
DeepSeek-V4-Pro is a cutting-edge text generation model developed by deepseek-ai, featuring 862 billion parameters. Updated one day ago, it offers advanced capabilities for generating human-like text across various applications.

Token Classification
openai/privacy-filter
⇧ 57k Downloads
The privacy-filter model by OpenAI is designed for token classification tasks, specifically to identify and filter sensitive information in text. With 1 billion parameters, it was last updated six days ago and aids in enhancing data privacy.

Image-Text-to-Text
Qwen/Qwen3.6-27B
⇧ 508k Downloads
Qwen3.6-27B is an image-to-text model developed by Qwen, boasting 28 billion parameters. Updated four days ago, it excels in generating descriptive text from images, facilitating applications in image captioning and understanding.

Text Generation
deepseek-ai/DeepSeek-V4-Flash
⇧ 96k Downloads
DeepSeek-V4-Flash is another advanced text generation model from deepseek-ai, comprising 158 billion parameters. Updated one day ago, it provides efficient and high-quality text generation for various use cases.

Image-Text-to-Text
moonshotai/Kimi-K2.6
⇧ 500k Downloads
Kimi-K2.6, developed by moonshotai, is an image-to-text model with 1.1 trillion parameters. Updated five days ago, it delivers exceptional performance in generating textual descriptions from images, supporting diverse applications in computer vision and natural language processing.

TRENDING AI TOOLS

• 🦖 T-Rex Label: AI-powered data labeling tool for efficient and accurate annotations.

• 🧠 Braintrust: The observability and evals platform leading teams use to ship reliable AI

• 🔍 ultrareview: Comprehensive tool for automated code reviews and quality analysis.

• 🧠 GPT 5.5: Advanced AI model for complex tasks like coding, research, and data analysis.

print("Everything else")

• The Karpathy-Inspired Claude Code Guidelines offer principles to enhance Claude Code behavior, addressing common LLM coding pitfalls.

• Anthropic has surpassed OpenAI with a $1 trillion valuation, reflecting its rapid growth and increasing dominance in the AI industry.

• Researchers introduce Verbalized Sampling, a prompting strategy that makes models like GTP or Claude more creative.

• NotebookLM can now auto-label & categorize sources (when you have 5+), so you can spend less time scrolling and more time thinking/learning/philosophizing, etc.

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