Hi there, this is your daily ☕️ Techpresso.
In today's Techpresso:
🤖 Microsoft launches ‘vibe working’ in Excel and Word
👀 Nvidia CEO says China is nanoseconds behind US
⚛️ Caltech builds the world’s largest neutral-atom quantum computer
💥 US wants Taiwan to make half its chips in America
👨👩👧👦 OpenAI releases parental controls for ChatGPT
🚀 DeepSeek debuts new AI model as ‘intermediate step’ towards next generation
🎁 + 16 other news you might like
🔮 + 4 handpicked research papers and tools
🤖 Microsoft launches ‘vibe working’ in Excel and Word LINK
Microsoft introduces "vibe working" with an Agent Mode for Excel and Word on the web, letting you generate complex reports or draft articles by working iteratively with Copilot through simple prompts.
A separate Office Agent powered by Anthropic models now works inside Copilot Chat to build full PowerPoint presentations and research papers by asking clarifying questions and conducting web-based searches.
These new tools are currently online for Microsoft 365 Personal and Family subscribers, but the Excel feature requires installing a special Excel Labs add-in to function for now.
👀 Nvidia CEO says China is nanoseconds behind US LINK
Nvidia CEO Jensen Huang claims China is just nanoseconds behind the US in chipmaking and argues that America should continue selling its technology there to maintain its geopolitical influence.
Following export restrictions, the company is now shipping a compliant H20 AI GPU to Chinese customers, its second attempt to create a tailored processor after the A100 and H100 bans.
Meanwhile, Huawei is shipping systems with its Ascend 920B silicon and other firms are investing in custom designs to create a CUDA-free ecosystem, directly challenging Nvidia's previous market dominance.
⚛️ Caltech builds the world’s largest neutral-atom quantum computer LINK
Caltech physicists built the largest neutral-atom quantum computer by trapping 6,100 cesium atoms as qubits in a single array, a significant increase over past systems with only hundreds.
The system achieved coherence times of about 13 seconds, nearly 10 times longer than earlier experiments, while performing single-qubit operations on the atoms with an accuracy of 99.98 percent.
Using "optical tweezers," the team showed it could move individual atoms within the array without breaking their quantum state, a key feature for building future error-corrected quantum machines.
💥 US wants Taiwan to make half its chips in America LINK
The Trump administration is pushing Taiwan to relocate its semiconductor production so that 50% of the chips America needs are manufactured domestically to ensure supply-chain security for the country.
To enforce this move, the White House threatened steep tariffs and a “1:1” production rule, securing a purported $165 billion investment pledge from TSMC for new U.S. chip plants.
Transplanting the industry is a major challenge due to its complex global supply chain, with Taiwanese officials arguing that no single country can fully control the entire semiconductor manufacturing process.
👨👩👧👦 OpenAI releases parental controls for ChatGPT LINK
OpenAI now lets parents link their account to a teen's to manage core features like turning off 'model training', 'memory', 'voice mode', 'image generation', and setting 'quiet hours'.
While you cannot see your teen's conversations to respect their privacy, you will get notifications if the AI detects content that could pose a serious risk of harm.
The company also launched a new resource page for parents that explains how ChatGPT works, details the available controls, and offers tips on how teens can use AI safely.
🚀 DeepSeek debuts new AI model as ‘intermediate step’ towards next generation LINK
Chinese AI startup DeepSeek released an experimental model that debuts a technique called Sparse Attention, designed to improve efficiency when handling long sequences of text without losing output quality.
The new method uses a "lightning indexer" to selectively score and rank past tokens, allowing the system to focus only on the most relevant information for each specific query.
This approach results in 2–3 times faster inference for long contexts and cuts memory usage by 30–40 percent, while maintaining nearly identical performance on reasoning and coding benchmarks.
Other news you might like
- The Jaguar Land Rover Hack Is So Bad Automaker Needs a $2 Billion LoanLINK
- EA acquired by Saudi Arabian investment fund and others for $55 billionLINK
- X says will fight Indian court ruling on content takedown systemLINK
- EU pushes for Chips Act 2.0 investment as it looks set to miss global silicon production targets by a wide margin — seeks quadrupling of semiconductor investment as $50 billion initiative floundersLINK
- Apple is reportedly nearing production for its latest M5-powered MacBooksLINK
- Google’s gradient ‘G’ icon, design is going company-wideLINK
- Regulators struggle to keep up with the fast-moving and complicated landscape of AI therapy appsLINK
- Marissa Mayer will close her old AI startup, sell assets to her new AI startupLINK
- Pressure mounts on Siri as ChatGPT ads start airing on primetime TVLINK
- Tim Berners-Lee on putting the world wide web in the public domainLINK
- TSMC says Intel didn’t ask for investments — denies existence of talks for partnership, joint ventureLINK
- DJI loses lawsuit over classification as Chinese military companyLINK
- US-made Leonidas microwave weapon takes down 49 drones with a single blastLINK
- Deepmind says video models for visual tasks could become what LLMs are for text tasksLINK
- Ford doesn't think CarPlay Ultra is worth using now, but will look at future versionsLINK
- Meta has introduced revised guardrails for its AI chatbots to prevent inappropriate conversations with childrenLINK
Latest research and tools
DeepSeek-v3.2-Exp: an experimental language model designed to efficiently process long texts using a new sparse attention mechanism, without sacrificing performance compared to its previous version.LINK
Linus Learns Analog Circuits: a personal project that documents its creator's journey of learning how to design and build analog and mixed-signal electronic circuits, starting with guitar pedal kits as a hobby.LINK
C-sigma: a C library for creating proofs that you know a secret, like a password or private key, without having to reveal the secret itself to anyone.LINK
A Deep Research Agent for Curating Vision Datasets: this paper introduces an AI agent that automatically finds and fixes errors in large image datasets.LINK
Want to get the latest news differently? Find us on:
See you tomorrow for a new dose of ☕️ Techpresso!
.png){kind=small}
.png){kind=small}
.png){kind=small}