Listen to today’s podcast: https://www.youtube.com/channel/UC-nqwUyvLDEvs7bV985k-gQ
Welcome back to AI Daily. Today’s podcast episode was created from the following stories:
5 acquisitions, winning over skeptical engineers, and spending tens of millions: Inside a public company’s ‘AI native’ push
Source: businessinsider.com
Amplitude is racing to become an AI native analytics company after a major strategic pivot that included hiring an AI-centric engineering chief, running an all-hands AI week, and acquiring five startups since late 2024. The company says developer productivity rose about 40 percent, with some teams seeing multiples of that, as internal adoption of tools like Cursor and GitHub Copilot surged. It has invested tens of millions of dollars (possibly over 100 million) to retool its products and culture around AI.
Challenges and Research Directions for Large Language Model Inference Hardware
Source: arxiv.org
This research note argues that LLM inference is increasingly bottlenecked by memory capacity and interconnect, not raw compute. The authors outline promising paths forward, including high bandwidth flash, processing-near-memory with 3D stacking, and low-latency interconnects, with datacenter-first designs that could inform mobile. The takeaway: future speedups will come from moving data smarter, not just crunching faster.
USA: Die Architektur der Abschiebung und Palantirs Rolle im neuen ICE-System
Source: heise.de
ICE is modernizing deportation operations by expanding Palantir’s Gotham-based ICM platform with a deportation-focused ImmigrationOS and integrating data from brokers and surveillance tools like Penlink’s Webloc and the AI-assisted Tangles OSINT system. Field agents also use ELITE for mobile targeting and Clearview AI for face recognition, amid an exclusive 2026 contract that frames Palantir as operationally indispensable. The piece highlights escalating capabilities, centralization, and civil liberties concerns around large-scale data fusion in immigration enforcement.
Announcing Vortex Support in DuckDB
Source: duckdb.org
DuckDB now ships a core extension for Vortex, an open columnar format that supports compute on compressed data and late decompression, aiming to cut I/O and memory use. On TPC-H at scale factor 100, Vortex ran 18 percent faster than Parquet v2 (and 35 percent versus v1) in the authors’ tests, with more consistent runtimes across cold and warm runs. Designed for analytics and AI pipelines, Vortex targets efficient CPU/GPU paths and multimodal data layouts.
Google Drive’s new ‘smart’ AI features are forcing me to move all my private documents
Source: androidauthority.com
Google has enabled AI summaries in Drive by default for Gemini subscribers and Workspace users, prompting strong privacy concerns over scanning file contents without explicit opt-in. While Google says data is encrypted and not used to train models, it manages the keys and triggers summaries proactively, pushing some users to disable the feature or move sensitive files to client-side encrypted alternatives. The story underscores a growing tension between convenience and consent in consumer AI.
PORTRAIT – 5 QUESTIONS SUR EDWIN CHEN, LE MILLIARDAIRE DE L’IA DONT VOUS N’AVEZ JAMAIS ENTENDU PARLER
Source: presse-citron.net
Edwin Chen’s Surge AI built a lucrative niche supplying high-quality human feedback data to frontier model labs, prioritizing expert annotators and rigorous evaluation over low-cost bulk labeling. Self-funded and profitable, the company reportedly did 1.2 billion dollars in 2024 revenue with a valuation near 24–30 billion, making Chen one of the youngest ultra-wealthy founders. He warns the AI industry is over-optimizing for leaderboard status rather than real-world reliability and safety.
AI開発にグローバルサウスが異議を唱える──特集『THE WIRED WORLD IN 2026』
Source: wired.jp
Nighat Dad argues that AI’s boom rests on invisible, precarious labor in the Global South and the extraction of public data without meaningful consent, echoing a modern form of digital imperialism. She highlights emerging pushback: Brazil’s transparency mandates for training data, Kenya’s investment requirements, and grassroots efforts demanding fair wages, stronger safeguards, and accountable AI. The piece reframes AI governance as a global equity issue, not just a technical one.
Taken together, these stories trace the arc of AI from silicon to software to society: companies racing to operationalize AI at scale, researchers re-architecting hardware for inference bottlenecks, and communities and consumers demanding agency over how data and algorithms shape their lives. As adoption accelerates, the defining advantage will not just be speed or spend, but trust — built on transparent systems, responsible data practices, and measurable real-world value.
