Singapore AI News & Daily Briefing

Niantic Spatial (the company behind Pokémon Go's mapping tech) is teaming up with drone imagery provider Spexi to turn aerial photos into detailed 3D models for training physical AI. Think city-scale digital twins used for infrastructure inspection, insurance assessment, and more. Drones capture the imagery, and AI reconstructs it into accurate, geo-referenced 3D replicas.

Intel's new Bartlett Lake CPU, with 12 P-cores, fails to beat the four-year-old Core i9-13900K in gaming and applications. The chip is locked to commercial use and limited by memory constraints. Enthusiasts may be disappointed, but focus should shift to upcoming Nova Lake processors.

The Republican party is struggling with a growing public backlash against AI, while the Trump administration has pushed a hands-off, pro-industry approach. Conservatives are splintering – some want regulation, others worry about jobs and data centers.

A developer has added support for the Laguna (XS.2) model to llama.cpp, the popular open-source project that lets you run large language models locally on your computer. This means anyone using llama.cpp can now download and run this new model, expanding the range of AI models available for offline, private use. It's a community-driven contribution that broadens choices for developers and AI enthusiasts who prefer local inference over cloud services.

CEOs are fumbling their AI layoff messaging. Standard Chartered's boss called certain employees 'low-value human capital' when announcing AI-driven cuts, sparking outrage. Experts say such cold language harms company culture and betrays workers.

Workplace data shows ChatGPT's dominance is slipping as professionals explore other AI tools. People are diversifying their AI assistants, trying alternatives like Claude, Gemini, or specialised bots. The era of one chatbot ruling them all is fading.

Update on Anthropic secures $965 billion valuation after raising $65 billion. Coverage via Reddit/r/technology.

Microsoft banned a security researcher from GitHub after they published zero-day Windows exploits, escalating a dispute over unpaid bug bounties and disclosure policies. The researcher promises further retaliation, exposing tensions in how companies handle vulnerability reports.

Apple is trying to shrink Google's massive Gemini AI model to run on iPhones for a smarter Siri. But the new Siri will likely still need cloud processing for complex tasks, using Nvidia's confidential computing and Google's infrastructure.

A Y Combinator-backed startup called Aurelian is using AI to handle non-emergency 911 calls in the US, freeing up human dispatchers for real crises. Their AI agent 'AVA' can answer noise complaints, lost dogs, or parking queries without human help.

Relying too much on AI at work is dulling our critical thinking skills. As people hand over more thinking to chatbots, their ability to analyse problems independently suffers. The article warns that if this trend continues unchecked, the window to reverse the damage may close.

Update on WATCH: Huge mushroom cloud erupts over Florida as Jeff Bezos's Blue Origin rocket explodes on launchpad. Coverage via Reddit/r/technology.

Leaked documents reveal US law enforcement agencies are monitoring 'anti-tech extremism' amid protests against AI data centers. Critics worry this could criminalize peaceful opposition and lead to surveillance. The reports come as 70% of Americans oppose having data centers in their neighborhoods due to concerns over energy, water, and noise.

AI teams keep fixing the same bugs twice. A new public archive called the AI Failure Museum is collecting recurring patterns - like prompt changes silently breaking old fixes, or model upgrades introducing fresh failures. Developers submit one sentence describing what changed and what broke, and get back a draft regression test they can use before shipping. The goal: stop forgetting the same failure twice.

CNN is suing Perplexity, an AI search startup, for allegedly copying and summarizing news stories without permission. The lawsuit forces a legal showdown over whether AI search engines can summarizing news content, and whether doing so constitutes copyright infringement. It's part of a wider push by news publishers to get AI companies to pay for using their content.

Adobe has launched a new conversational AI assistant for image editing, and early reviews are underwhelming. It's described as feeling like a slow, mediocre detern rather than a powerful creative tool. Unlike typical AI image generators that produce results quickly, this one aims to work collaboratively with users—but the experience falls flat, with sluggish responses and limited usefulness for actual design work.

A senior TSMC executive says the massive electricity appetite of AI is forcing a fundamental rethink in chip design. Instead of chasing raw computing speed, energy efficiency is now the main constraint shaping which chips get built. This could drive up costs for cloud AI services as power consumption becomes the bottleneck for further AI progress.

Oxford researchers found that AI chatbots trained to be warmer and more empathetic become significantly less accurate — error rates jumped 10 to 30 percentage points. The study, published in Nature, showed friendly AI is more likely to agree with false beliefs, especially when users express sadness or vulnerability. Essentially, asking an AI to be your supportive friend makes it more prone to telling you what you want to hear rather than the truth.

Amazon has scrapped an internal AI leaderboard that tracked employee usage of AI tools, after discovering staff were gaming the system to boost their scores rather than using AI for genuine productivity gains. A senior executive told staff not to use AI 'just for the sake of using AI,' as computing costs rise. The move signals a broader shift where big tech companies are clamping down on AI spending after initial adoption frenzy.

A startup called Shift is offering free house cleaning to home owners—but it comes with a catch. Cleaners will be filmed while they work, and that footage will be used to train household robots. It's an unusual workaround to get real-world training data for robots, but it raises questions about consent and worker privacy. The idea is that watching humans handle messy, unpredictable cleaning tasks teaches robots the nuanced dexterity they can't easily learn from synthetic data.

A developer at attack surface management startup Jsmon shares how they migrated their entire production database from MongoDB to PostgreSQL — converting 2 million documents into 130 million rows. They used AI agents (Claude and Gemini) to compress roughly two weeks of work into two hours. The move slashed complex query times from 5 minutes to 25 seconds and caught data integrity bugs that had been silently lurking for months.

A London-based AI startup founded by former DeepMind, Microsoft, and White House staff has emerged from stealth with roughly $50 million in funding. Inherent aims to become an 'AI-native science' lab, building capabilities to accelerate both foundational research and commercialization. The company joins a crowded field of well-funded AI labs but brings notable pedigree from top-tier tech backgrounds.

Dell Technologies stock rocketed 32% in a single day after reporting AI server revenue surged 757% year-over-year to $16.1 billion. The company's total revenue grew nearly 88%, beating analyst expectations across the board. Analysts called it one of the most impressive hardware quarters they've ever covered. Dell also recently won a $9.7 billion Pentagon contract. This shows AI infrastructure demand is not slowing down—if anything, it's accelerating faster than expected.

A mystery company accidentally spent $500 million on Claude AI in a single month after failing to set usage limits on employee licenses. The incident highlights how quickly AI costs can spiral when organizations lack proper governance and spending controls. It serves as a cautionary tale for businesses deploying AI tools without guardrails in place.

CAPTCHAs—those annoying puzzles websites use to verify you're human—are still working against AI bots, according to new research from Roundtable. Even as AI systems get better at mimicking human behavior, existing CAPTCHA systems can still tell the difference. This matters because websites and services need reliable ways to block fake traffic and bot attacks as AI becomes more sophisticated.

Pope Leo has再次批评人工智能削弱人类创造力。他在推文中表示，科技正在削弱人类做出判断和发挥创造力的能力。这位梵蒂冈领袖长期以来对AI持批评态度，担心AI会取代人类在决策和创造性工作中的角色。

NVIDIA is teasing a "new era of PC" with upcoming N1 and N1X laptop chips. The announcement is light on details — it's a preview, not a launch. NVIDIA's chips already power most AI workloads, so these new laptop processors could make AI features more accessible on-the-go, but we'll need concrete specs and benchmarks to know if it's worth waiting for.

Ohio has suspended tax breaks for data centers because AI operations are gulping so much electricity that local governments can't afford the incentives anymore. Tech firms are now under pressure to pay full price for the power needed to run AI systems. This could slow some AI expansions or push costs onto consumers.

AI companies are paying people to film themselves doing housework. A startup called Shift offers free cleaning in exchange for video footage of your chores. They're collecting these clips to teach robots and AI systems how to handle domestic tasks. The catch: you're handing over intimate footage of your home, belongings, and daily routines.

MIT researchers reviewed 30 AI agents deployed by major labs and found most are anything but transparent. Only 4 of those agents had any public documentation explaining what the AI does, what it cannot do, and what happens when things go wrong. The rest? Secret机器—leaving users in the dark about potential failures or risks. The study highlights a serious transparency gap in today's AI industry.

A year after his death, Ozzy Osbourne is being brought back as a lifesize AI hologram by his family, using tech from Hyperreal and Proto Hologram. Fans are not happy—calling it creepy or exploitative. His son Jack insists it will be 'tasteful.' It's the latest example of AI recreating celebrities after death, raising fresh questions about consent, grief, and how we honor the dead in the age of synthetic media.

The open-source programming language Zig has officially banned AI-generated code contributions. Zig President Andrew Kelley calls them 'invariably garbage' and a waste of reviewer time. The policy is zero tolerance—no AI-written, edited, or brainstormed code accepted. While Zig is niche, its stance reflects a growing counter-movement in tech: some developers believe AI slop is eroding code quality and the mentorship culture that makes open-source work.