Biggest data moments past 2 weeks: AI rivals face off, Musk’s $97.4B gamble, DeepSeek’s breakthrough, DBS’s AI-powered future

???Welcome to this issue!

AI is moving fast! From regulatory battles to major breakthroughs in AI infrastructure. Here’s everything you need to stay ahead.

?? AI’s biggest moves this month

A divided future at the AI Paris summit ??

The AI Action Summit in Paris made one thing clear: the world is far from agreeing on how to handle AI. The U.S. came in strong, with Vice President JD Vance making it clear they won’t let regulations slow them down, especially when it comes to competing with China. Meanwhile, Europe pushed for more responsible AI governance, but attempts at a global agreement fell apart. Beyond politics, big concerns took center stage: AI’s massive energy consumption, security risks, and the looming impact on jobs.

Dario Amodei, CEO of Anthropic, warned that by 2026-2027, AI could be as disruptive as “a country of geniuses in a data center,” reshaping industries and national security. While innovation is accelerating, global cooperation on AI safety remains uncertain. The road ahead? More urgency, clearer policies, and a balance between progress and protection.

Musk’s $97.4B power play ??

Elon Musk has launched a $97.4B bid to take over OpenAI’s nonprofit arm, challenging its shift to a for-profit model. Backed by investors like xAI and 8VC, Musk claims OpenAI has strayed from its mission. While CEO Sam Altman dismissed the offer with a sarcastic tweet, the move disrupts OpenAI’s transition—exactly what Musk intended.

Read Musk’s letter of intent here.

GitHub copilot gets smarter ??

GitHub previewed Agent Mode, enabling Copilot to fix its own errors, infer missing tasks, and iterate on code autonomously. Also launching: Copilot Edits in VS Code for AI-driven multi-file changes, new AI models, and next edit suggestions to speed up development. Plus, Project Padawan hints at a future where AI handles coding tasks end-to-end.

?? Breaking

?? DeepSeek’s AI breakthrough – Native Sparse Attention (NSA) makes long-context AI models faster & cheaper. But U.S. regulators are already investigating DeepSeek over security risks.

?? Musk’s xAI launches Grok 3 – Elon Musk’s new chatbot, ****Grok 3, is now live for Premium+ users on X.

?? Federal workers sue Musk’s DOGE agency – Over 100 current and former federal workers have sued Elon Musk and DOGE for allegedly accessing sensitive personnel records without proper authorization.

?? Industry insights

???The shift toward composable data platforms

The future of data is composability, and open table formats (OTFs) like Apache Iceberg, Delta Lake, and Apache Hudi are paving the way. These formats introduce table virtualization, allowing the same table to be accessed across multiple platforms without data duplication. This approach solves the long-standing challenge of data gravity, where data is locked within a single platform, making cross-platform collaboration difficult.

A major shift is happening with Confluent and Databricks’ new strategic partnership, which creates a bidirectional data flow between operational and analytical platforms. Using Confluent’s Tableflow and Databricks Unity Catalog, Kafka topics will appear as Delta tables, and Delta tables will be exposed as Kafka topics—seamlessly bridging real-time streaming with batch analytics. This marks a new era of interoperability, where platforms plug together like Lego bricks, eliminating rigid ETL pipelines and unlocking a truly composable data ecosystem.

??? Fluss: The streaming storage revolution

Real-time data needs speed, efficiency, and scale, and that’s where Fluss steps in. Positioned as an alternative to Kafka, Fluss uses columnar storage (Apache Arrow IPC) to boost read performance and cut network costs. The real game-changer is, it merges streaming and Lakehouse storage, eliminating redundant data layers and ensuring instant access to both real-time and historical data.

One standout feature is Delta Join, which makes Flink Stream-Stream Joins massively more efficient, reducing resource costs by 10x and eliminating huge state storage overhead. With Kafka compatibility and deeper Apache Flink/Iceberg integration on the horizon, Fluss is shaping up to be a serious contender in the real-time analytics space.

?? How DBS became an AI-powered bank

DBS, Singapore’s largest bank, didn’t just adopt AI, it transformed into a tech-driven institution. With a vision to operate like a startup, DBS focused on embedding AI and data into every part of the business. A major breakthrough was reducing AI deployment time from 18 months to under 5 months, thanks to ALAN, DBS’s industrialized AI platform.

AI now powers 45 million hyper-personalized customer interactions per month, drives fraud detection, HR insights, and operational efficiency, and has generated S$150M in additional revenue.At the core of this transformation was a culture shift, from training 8,000 employees in data literacy to gamifying AI learning with AWS Deep Racer. By breaking silos and running the bank through 60+ customer-centric AI-powered journeys, DBS turned itself into a truly digital-first bank.

?? Binary cross-entropy matters in ML

Binary cross-entropy (BCE) is a key loss function for binary classification tasks like spam detection and medical diagnosis. It not only measures prediction accuracy but also penalizes overconfidence in wrong predictions, improving model reliability. BCE integrates well with neural networks but struggles with imbalanced datasets and numerical instability when predictions approach 0 or 1. Solutions like weighted BCE and clipping probabilities help address these issues. Widely used in TensorFlow and PyTorch, BCE remains essential for deep learning models.

?? Tool of the fortnight: Here’s a great YouTube video that breaks down how LLMs like ChatGPT work, from tokenization to inference. ??

?? Pro tip: Deepnote’s flexible app layout is here! Drag, drop, and resize blocks to design stunning, multi-column data apps your way. See it in action!

? Final thoughts

From Silicon Valley power plays to game-changing data innovations, the last two weeks were packed with action. As we move forward, expect more disruption, more breakthroughs, and more ways to stay ahead. See you next issue!