All AI News

Researchers from the Allen Institute for AI and UC Berkeley have developed a new mixture-of-experts model called EMO that achieves near-full performance using only 12.5 percent of its experts. This breakthrough could make large AI models practical for memory-constrained devices for the first time. The innovation centers on training experts to specialize in content domains rather than generic word types.

The cybersecurity landscape is undergoing a fundamental transformation as artificial intelligence disrupts traditional training methods. According to recent analyses, the once-thriving Capture The Flag scene is struggling to adapt to AI-powered challenges. This shift is creating new dynamics in security testing and professional development.

NVIDIA has introduced its Cosmos World Foundation Model platform, representing a significant leap in physical AI development. This initiative includes advanced world models capable of generating realistic, controllable video content from real-world data.

A significant gap exists between the rapid adoption of new AI models like Codex and the utilization of their workflow features in data science and life sciences teams. This friction hinders the translation of AI capabilities into reproducible scientific discoveries and measurable research acceleration.

Researchers have developed a novel Mixture-of-Experts AI model that retains nearly full performance using only 12.5 percent of its specialized components. This breakthrough, called EMO, could make large AI models practical for memory-limited environments like smartphones and edge devices.

Databricks is integrating GPT-5.5 into its enterprise agent workflows following the model's record-breaking performance on the OfficeQA Pro benchmark. This move signals a significant advancement in AI-powered enterprise automation and reasoning.

The Repowise platform is revolutionizing how AI agents understand complex codebases through automated documentation and dependency analysis. By generating structured wikis and architectural graphs in under 30 seconds, developers can provide AI assistants with accurate codebase context. This open-source tool bridges the gap between AI capabilities and practical software engineering needs.

The unprecedented AI-driven bull market has not triggered the typical flood of new stock offerings, a phenomenon known as de-equitisation. This suggests a fundamental shift in corporate finance strategy, where firms are leveraging AI efficiencies to fund growth internally, potentially ending the era of easy equity issuance.

Prominent Bollywood stars, including global icon Aishwarya Rai Bachchan, are leading a high-profile fight against AI-fuelled identity theft. Their personal cases are becoming catalysts for new legislation aimed at curbing the surge of fake online content. This marks a critical legal frontier for celebrity rights in the digital age.

The traditional mortgage industry is undergoing a seismic shift as technology startups introduce digital solutions. These innovations aim to streamline the cumbersome home loan process and enhance customer experience. The role of the traditional mortgage broker is being fundamentally challenged.

Japanese flash memory maker Kioxia is reporting a staggering surge in profits, driven by soaring demand for AI hardware. The company, a Toshiba spinout, is now planning to list American depositary shares to attract a wider pool of US investors.

Law firms are increasingly developing proprietary AI tools to streamline operations and enhance client services, creating new revenue streams. This trend marks a shift from buying generic software to building tailored solutions that address specific legal workflows and client needs.

New research reveals that measuring AI's impact on productivity requires a fundamental shift from tracking activity to quantifying real output. Self-reported speed estimates are proving inadequate for assessing genuine efficiency gains. Experts argue that meaningful productivity metrics must focus on tangible results rather than time saved.

A new documentary featuring former Google executive Mo Gawdat delivers an insider warning on the ethical development of artificial intelligence. The film, 'Chasing Utopia,' takes a measured yet urgent look at the potential risks and necessary safeguards for a tech-enhanced future.

Nio's new Onvo L80 electric SUV enters pre-sale, aiming to address unmet needs in the five-seat segment. The vehicle boasts a cavernous 2,840 liters of maximum storage space and a range of 605 kilometers, positioning itself as a practical family EV.

AI chipmaker Cerebras has made a historic market debut, with shares surging 68% following its highly anticipated IPO. The company's valuation now stands at approximately $95 billion, marking one of the most successful tech offerings in recent years. This explosive performance underscores intense investor appetite for artificial intelligence infrastructure.

A novel custom NVENC encoder bridge enables splitting large AI models like FLUX 2 across multiple GPUs over standard Ethernet or Wi-Fi, bypassing the need for expensive NVLink hardware. The technique dramatically accelerates image generation and promises to democratize high-performance AI workloads.

The landmark Musk v. Altman trial concluded its final week with attorneys fiercely attacking the credibility of both tech titans. Sam Altman defended against accusations of self-dealing while painting Elon Musk as a power-seeker who wanted overwhelming control of OpenAI. The jury must now decide which narrative to believe.

OpenAI has begun previewing new personal finance tools for ChatGPT Pro subscribers, enabling direct connections to bank and brokerage accounts via Plaid. This feature allows for personalized financial advice and dashboard management. The company plans to eventually roll out the functionality to all users.

YouTube is expanding access to its AI-powered likeness detection technology, previously reserved for celebrities and talent agencies, to all users over 18. The tool uses a selfie-style scan to monitor the platform for unauthorized deepfakes. This marks a significant escalation in the platform's fight against synthetic media.

A new course from DeepLearning.AI teaches practical AI prompting techniques to move beyond generic outputs. According to experts, providing specific context is the key to unlocking more useful and creative responses from generative AI systems. The course, led by Andrew Ng, aims to make advanced prompting accessible to all users.

A new approach to building AI agent systems uses dynamic tool exposure and context injection to dramatically improve efficiency. By exposing only necessary tools and injecting ephemeral context per step, these systems reduce computational costs and improve task accuracy. This methodology represents a significant shift from monolithic, always-on agent architectures.

A newly discovered flaw in the Linux kernel, the fourth this month, poses a significant risk of SSH host key theft. While a patch has been developed, it is not yet available for all distributions, leaving systems vulnerable. This incident follows a pattern of sophisticated attacks targeting core Linux infrastructure.

The preprint server arXiv is implementing a strict new policy to ban researchers who submit papers containing AI-generated slop. The policy targets papers with incontrovertible evidence of unchecked LLM output, such as hallucinated references. This move comes amid a growing wave of AI-generated content and new tools designed to detect citation hallucinations.