Homechevron_rightArchives

Archives

Explore Stories by Date

24 Mar 2026

Telling AI Models They're Experts Actually Makes Them Worse at Knowledge Tasks

University of Southern California researchers found that popular expert persona prompts harm AI model accuracy on factual tasks. While instructing AI to act as an expert helps with writing and safety, it degrades performance on math and coding by shifting models into instruction-following over factual recall mode. A new technique called PRISM aims to solve this tradeoff.

3 Sources

Norway wealth fund plans AI investment decisions with human oversight, CEO Tangen confirms

Norway's $2.1 trillion sovereign wealth fund will eventually allow AI systems to make some investment decisions under human supervision, though not yet due to accuracy concerns. Currently, half of the fund's 700 employees use Anthropic's Claude to code AI tools for monitoring 7,000 companies, simulating negotiations, and reducing transaction costs. CEO Nicolai Tangen reports billions in benefits from millions invested in AI.

2 Sources

TECNO launches EllaClaw, the first globally available OpenClaw AI agent for smartphones

TECNO has unveiled EllaClaw, the first globally available AI agent that integrates OpenClaw directly into smartphone operating systems. Unlike ChatGPT and other chatbots, EllaClaw can autonomously navigate multiple apps to complete complex tasks using natural language commands. The beta version promises consumer-ready features with built-in privacy safeguards for emerging markets worldwide.

3 Sources

Oracle's AI agents can decide and act autonomously, but who's liable when they fail?

Oracle unveiled 22 Fusion Agentic Applications that autonomously reason, decide, and act across finance, HR, supply chain and customer experience workflows. The shift marks a move from passive systems of record to active systems of outcomes. But analysts warn of unresolved liability questions and data integration challenges that could hinder adoption.

11 Sources

Alibaba unveils XuanTie C950 RISC-V chip designed for AI agents and cloud workloads

Alibaba revealed its XuanTie C950, claiming it's the most powerful RISC-V server chip available. Built on 5nm process technology, the processor runs at 3.2 GHz and is optimized for AI workloads including agentic AI and large language models. However, performance benchmarks suggest it trails western chips by several years, with capabilities comparable to Apple's 2020 M1 chip.

3 Sources

Google Gemini lets you import chat history and memories from rival AI chatbots

Google launched new switching tools that allow users to transfer memories and entire chat histories from competing AI chatbots directly into Gemini. The memory import feature helps users share key preferences, relationships, and personal context without retraining the AI assistant. Available for free and paid consumer accounts, the move aims to help Google catch up with ChatGPT's 900 million weekly active users.

26 Sources

MoonPay launches open-source wallet standard to help AI agents manage funds across blockchains

MoonPay has released an open-source wallet framework designed to let AI agents hold funds and execute transactions across multiple blockchains. The Open Wallet Standard (OWS) addresses wallet and key management fragmentation while keeping private keys secure. Major contributors include PayPal, Ethereum Foundation, Coinbase, Stripe, and OKX.

2 Sources

Broadcom sounds alarm as TSMC capacity constraints threaten AI chip supply through 2026

Broadcom has flagged critical supply constraints across the semiconductor industry, with TSMC hitting production capacity limits as soaring demand for AI chips strains global manufacturing. The bottleneck extends beyond semiconductors to lasers and printed circuit boards, prompting companies to secure multi-year supply agreements stretching three to four years.

3 Sources

Microsoft-backed Lace Lithography raises $40M to create chip designs 10 times smaller

Norway-based Lace Lithography has secured $40 million in Series A funding to develop helium atom beam lithography technology. The Microsoft-backed startup aims to create chip designs 10 times smaller than what's currently possible, potentially transforming semiconductor manufacturing with atomic resolution capabilities by 2029.

3 Sources

Alibaba launches Accio Work agentic AI platform for small and medium-sized businesses

Alibaba International has unveiled Accio Work, an agentic AI platform designed to help small and medium-sized businesses automate complex operations without coding. The plug-and-play system deploys AI agents to handle compliance, marketing, and sourcing while drawing from real-time consumer trends to minimize AI hallucinations.

2 Sources

ZYT's AI autonomous driving system already outperforms its CEO on crowded Shenzhen streets

Chinese autonomous driving startup ZYT is set to unveil a mobility foundation model at Beijing's auto show in April that CEO Shen Shaojie admits drives better than he does. The DJI spin-off trained its AI on diverse video sources including drones and vacuum cleaners, creating a system that adapts across vehicle types without dedicated modules for detection.

3 Sources

AI child abuse content surges 260-fold as watchdog warns of escalating criminal misuse

The Internet Watch Foundation identified 8,029 AI-generated child sexual abuse images and videos in 2025, marking a 14% annual increase. More alarming is the 260-fold surge in videos, with 65% classified as the most extreme category. The findings expose how generative AI tools enable bad actors to create violent, realistic abusive content with minimal technical skill, intensifying pressure on governments to regulate unregulated AI companies.

3 Sources

Delve halts demos as compliance allegations trigger investor retreat and fraud scrutiny

Y Combinator-backed Delve, valued at $300 million, faces serious fraud accusations after a whistleblower claimed the AI compliance startup fabricated certifications. The company has disabled demo bookings while investor Insight Partners scrubbed its $32 million investment announcement, signaling potential damage control as the scandal unfolds.

2 Sources

Meta Acquires Dreamer Founding Team to Accelerate Personalized AI Agents Development

Meta has completed an acqui-hire of AI startup Dreamer, bringing aboard tech veterans David Singleton, Hugo Barra, and Nicholas Jitkoff to its Superintelligence Labs. The deal includes non-exclusive licensing of Dreamer's technology, which helps users build personal AI agents. Dreamer remains a standalone entity while its founders join Meta's aggressive push into agentic AI.

6 Sources

23 Mar 2026

Claude AI can now control your computer to complete tasks, but security risks remain

Anthropic has enabled Claude Code and Claude Cowork to take direct AI computer control of MacOS desktops, allowing the tools to point, click, and navigate screens to complete tasks. Available as a research preview for Claude Pro and Max subscribers, the feature lets AI agents to complete tasks by opening files, using browsers, and running dev tools—though Anthropic warns it's error-prone and raises security concerns.

33 Sources

Get all updates into your inbox 💌, Stay Up-to-date for Free 🥳

Your Daily Dose of Curated AI News

Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Spend less time searching for the latest in AI and get straight to action.

The Outpost

News

About