Tuesday 25 March 2025 | Join Free | Upgrade

Together with Sponsor

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🔄 OpenAI reshuffles leadership as Sam Altman pivots to technical focus

🧠 Alibaba unveils its latest AI model "Qwen2.5-VL-32B"

💥 Chinese robot maker aims to match Tesla's 5,000 humanoids this year

💰 Meta faces $1B EU fine over ‘pay or consent’ model

🏢 Alibaba chairman warns of AI data center bubble

🚫 Binance fires staff and bans market maker amid trading scandal

🎁 + 9 other news you might like

🔮 + 5 handpicked research papers and tools

FROM OUR PARTNER


Be the First to See What’s New in Apache Airflow® 3.0

Airflow 3.0 is almost here and it’s the biggest release yet for the world’s most widely used data orchestration tool.

Join Astronomer live on April 23, 2025 at 11am ET for a deep dive into this game-changing release, packed with the most requested features from the Airflow community including:

  • A new UI (yes, including dark mode 😎)

  • DAG versioning, backfills, and a new asset-centric syntax

  • Remote execution and enhanced task isolation

Led by Airflow contributors and experts from Astronomer, this session will include a technical demo, upgrade tips, and live Q&A to help you prepare for a smooth transition to 3.0.

Register now

🔄 OpenAI reshuffles leadership as Sam Altman pivots to technical focus LINK
  • OpenAI is reorganizing its leadership team, with COO Brad Lightcap taking on expanded business responsibilities while CEO Sam Altman shifts his focus toward AI development and product strategy.
  • The restructuring accompanies OpenAI's evolution from a research organization to a commercial AI company, with recent appointments including Mark Chen as Chief Research Officer and Julia Villagra as Chief People Officer.
  • This leadership transition occurs as OpenAI diversifies its infrastructure partnerships beyond Microsoft, secures major investments from SoftBank, and develops new enterprise AI offerings to compete in the market.
🧠 Alibaba unveils its latest AI model "Qwen2.5-VL-32B" LINK
  • Qwen2.5-VL-32B, the new language model from China, offers improved capabilities including better human preference alignment, enhanced mathematical reasoning, and superior visual analysis compared to its predecessors.
  • Benchmark results show Qwen2.5-VL-32B outperforming competitors like Gemma 3-27B, Mistral Small 3.1 24B, and a specific version of GPT-4o, positioning it as a strong contender in the AI landscape.
  • At 32B parameters, the model strikes an optimal balance between powerful GPT-4-class capabilities and reasonable resource requirements, allowing users to run it alongside other applications on 64GB systems.
💥 Chinese robot maker aims to match Tesla's 5,000 humanoids this year LINK
  • Chinese robotics start-up Agibot plans to produce between 3,000 and 5,000 humanoid robots this year, matching Elon Musk's Optimus project target while expanding with a new plant in Shanghai's Pudong district to boost monthly capacity to over 400 units.
  • Agibot launched the Genie Operator-1 (GO-1) AI model in March 2025, which processes extensive image and video data to enhance humanoid robots' ability to interpret human actions, improving task completion rates from 46% to 78% in recent tests.
  • The company's flagship Yuanzheng A2 robot stands 175 cm tall, weighs 55 kg, and features AI-powered sensors that process multiple data types, while Agibot predicts household adoption of humanoid robots will take approximately five more years.
💰 Meta faces $1B EU fine over ‘pay or consent’ model LINK
  • The European Commission is preparing to fine Meta up to $1 billion for allegedly violating the Digital Markets Act with its "pay or consent" model, which regulators claim restricts consumer choice by forcing users to either pay for ad-free service or accept data tracking.
  • Meta faces potential penalties of up to 10% of its global annual revenue if found guilty, with CEO Mark Zuckerberg and U.S. officials strongly objecting to what they view as excessive European regulatory actions against American tech companies.
  • This potential sanction follows previous EU penalties against the company, including a €797.72 million fine for tying Facebook Marketplace to its social media platform and a record €1.2 billion GDPR fine for inadequate data transfer protections.
🏢 Alibaba chairman warns of AI data center bubble LINK
  • Alibaba Chairman Joe Tsai has expressed concern about potential over-investment in AI data centers, suggesting a bubble may be forming as companies plan to spend hundreds of billions of dollars on infrastructure ahead of actual demand.
  • Despite his caution, Alibaba itself is committing approximately $52 billion to AI and cloud computing over the next three years, with CEO Eddie Wu emphasizing their pursuit of artificial general intelligence to expand technological boundaries.
  • The tech industry's massive AI investment trend continues globally, with America's four largest technology corporations expected to allocate at least $320 billion to artificial intelligence initiatives this year alone.
🚫 Binance fires staff and bans market maker amid trading scandal LINK
  • Binance suspended an employee for front-running trades using privileged information about upcoming token launches, specifically purchasing Movement's MOVE tokens before their official listing and selling them for early profits.
  • The cryptocurrency exchange also banned a market maker who engaged in manipulative trading by dumping approximately 66 million MOVE tokens one day after launch, generating around $38 million in profits through unethical practices.
  • Movement Network has committed to compensating affected users through a token buyback program utilizing the $38 million recovered from the frozen market maker's earnings, with repurchased tokens being transferred to the Movement Strategic Reserve.

FROM OUR PARTNER

Sponsor

Innovating with AI’s new program, AI Consultancy Project, trains AI enthusiasts with all the skills and strategies they need to capitalize on the rapidly growing AI consulting market – which is set to 8x by 2032.

Some of the highlights current students are excited about:

  • The tools and frameworks to find clients and deliver top-notch services
  • A 6-month plan to build a 6-figure AI consulting business
  • Students getting their first AI client in as little as 3 days
👉 Click here to request early access to The AI Consultancy Project 👈

Other news & articles you might like

  • Waymo plans to bring its driverless taxis to Washington in 2026 LINK
  • Napster pioneered music sharing over 25 years ago. It just got bought for $207 million LINK
  • Europe is looking for alternatives to US cloud providers LINK
  • A Chinese tech giant says it slashed AI costs using only Chinese chips LINK
  • Chinese hackers spent four years inside Asian telco’s networks LINK
  • Microsoft launches AI agents to automate cybersecurity amid rising threats LINK
  • You can now download the source code that sparked the AI boom LINK
  • OpenAI says its AI voice assistant is now better to chat with LINK
  • a16z- and Benchmark-backed 11x has been claiming customers it doesn’t have LINK

Latest research and tools

VGGT: a feed-forward neural network that reconstructs 3D scenes by inferring camera parameters, depth maps, and point tracks from multiple images within seconds. LINK
Gatehouse: a composable authorization framework in Rust that allows combining multiple policy types (RBAC, ABAC) with logical operators to determine access permissions based on customizable criteria. LINK
Hann: a fast approximate nearest neighbor search library for Go that enables efficient similarity searching in high-dimensional spaces through various indexing methods like HNSW, PQIVF, and RPT. LINK
End-to-End Framework for Design Space Exploration of Microfluidic-Cooled 3DICs: researchers propose a comprehensive framework for exploring design options for three-dimensional integrated circuits cooled by microfluidic technology. LINK
Accurate INT8 Training Through Dynamic Block-Level Fallback: researchers present a method that enables efficient INT8 quantization for neural network training while maintaining accuracy by dynamically switching to higher precision for problematic blocks. LINK

Want to get the latest news differently?

Not subscribed to ☕️ Techpresso yet? Subscribe for free