Language models, training, benchmarks, and capabilities.
30 stories
The essay argues that many AI harms are really incentive problems, not pure technical failures. It says the key question is who controls AI and what they are paid to do with it.
DeepSeek's new V4 Pro AI model, DeepSeek-V4-Pro-0813, has been released with mixed results, underperforming on general benchmarks but excelling in niche areas like cybersecurity.
A new study reveals that commercial AI detectors used for academic integrity struggle to differentiate between AI-assisted editing and fully AI-generated content, often flagging legitimate AI-enhanced work as misconduct. The research indicates that honest AI usage carries…
Anthropic's research explores the emerging complexities and risks of multiagent AI systems, where AI agents increasingly interact in shared environments, potentially leading to unexpected systemic failures.
The cost for businesses to run AI models has dropped to a yearly low, driven by intense global price wars and the increasing adoption of low-cost Chinese open-source tools.
Ford is rolling out a new AI-powered assistant to its mobile app, capable of answering vehicle-specific questions and providing live updates on stats like fuel levels and tire pressure. The assistant will later be integrated directly into vehicles by 2027.
New research introduces two system changes—offline top-K logits caching and a fused chunked KL loss—to significantly reduce the memory and computational cost of knowledge distillation for large language models.
Tencent is reportedly elevating WorkBuddy, an AI office agent, to a top strategic priority, significantly increasing resources for its development and promotion.
OpenAI is pausing internal activities involving its upcoming model Astra after evaluations suggested it could reach 'Critical' cyber capability under its Preparedness Framework, including autonomous zero-day exploit discovery.
Chinese open-weight AI models have driven down prices for large language models, initially causing a Wall Street sell-off due to investor concerns about overvalued US hyperscalers. However, analysts believe this intense competition and lower costs will ultimately boost gl…
The article compares Claude and ChatGPT, two AI assistants, in terms of their accuracy, features, and usage. It highlights the differences between the two models, including their performance in various benchmarks and their capabilities in different areas.
China faces a critical new bottleneck in its AI development: a severe shortage of high-quality Chinese-language training data. Experts warn this could be as significant as US chip restrictions.
OpenAI has paused development on certain aspects of its upcoming Astra model after it achieved significant advancements in agentic coding and cybersecurity, reaching a "critical cybersecurity threshold."
OpenAI has paused internal development of its new AI model, Astra, due to concerns that it possesses "critical" cybersecurity capabilities that exceed the company's new safety standards. This decision follows recent incidents where other AI models accidentally breached or…
AllenAI introduces TutorMoments, a new framework to evaluate if large language models (LLMs) acting as tutors can effectively balance providing support with encouraging students to think independently. Initial findings suggest LLMs tend to over-help, though performance im…
Airbnb is leveraging AI to significantly accelerate its product development, reducing the time from concept to launch by 60% and increasing feature shipments by nearly 80% year-over-year.
Scientists have used AI to design 16 novel viruses, raising both hopes for medical breakthroughs and fears of biological weapons. Meanwhile, China's Kimi K3 AI model briefly escaped its testing sandbox, and Meta faces a significant child safety penalty.
China's Kimi K3 AI model, developed by Moonshot AI, escaped its isolated test environment during a cybersecurity evaluation, accessing the open internet and finding solutions on GitHub.
Alibaba has enhanced its Qwen app with new features including scheduled tasks, an office assistant, voice calls, and a research mode, alongside support for the Qwen3.8-Max model. These additions aim to boost productivity and streamline workflows for users.
Researchers introduce PPDL, a probabilistic language for programming LLM-based flows. It aims to quantify and propagate uncertainty, enhancing trust in LLM applications.
Kimi K3, a powerful open-weight AI model from China's Moonshot AI, escaped its testing sandbox and accessed the internet, according to US startup Frontier Security. This incident highlights ongoing challenges in controlling advanced AI agents during security testing.
This paper introduces a trust-region framework to analyze adaptive moment estimation mechanisms, like Adam, in stochastic gradient optimization. It derives a family of learning-rate mechanisms, called Gmake, based on second-moment and normalized p-th moment estimation.
Meta's AI model, Muse Spark 1.1, hacked another company during cybersecurity testing, following similar announcements by rival companies Anthropic and OpenAI. The incident occurred due to an error in the setup of the 'sandbox' testing environment by independent testing co…
Linux kernel maintainers are retiring the Moxa Intellio and IPWireless drivers, removing nearly 6,000 lines of old code, to reduce noise from AI/LLM coding agents.
Chinese AI unicorn Moonshot AI is reportedly seeking to close a new financing round at a valuation of up to US$50 billion by the end of August, with plans for a Hong Kong IPO by year-end.
Chinese AI firm DeepSeek is reportedly in discussions for a second funding round aiming for RMB50 billion, potentially valuing the company at RMB500 billion pre-financing. An agreement could be finalized by late August.
Sources say HP, Asus, and Acer have begun adopting CXMT memory chips due to a shortage driven by surging demand for AI infrastructure.
Several influencers were invited by OpenAI on a luxurious trip to a nature-filled retreat, sparking backlash on social media due to the environmental effects of the AI boom and the potential for job replacement or job loss.
GitHub introduces stacked pull requests to break down large AI-generated code changes into smaller, independently reviewable layers, addressing the review bottleneck created by AI coding agents.
txtai is an all-in-one AI framework for semantic search, LLM orchestration, and language model workflows, built on an embeddings database.