Cactus-Compute has launched Whistle, a highly efficient 16.9 MB speech recognition model designed to run locally on CPUs without external dependencies. The lightweight model supports multiple languages and provides transcription, word-level timestamps, and speech embeddings. This release highlights a growing trend in highly optimized, on-device AI tools alongside other recent efficiency breakthroughs like Tencent's STEPQuant framework.
Samsung's preliminary third-quarter results project a massive 780% surge in operating profit, reaching $80.2 billion (over 100 trillion won). This historic financial milestone makes Samsung the first tech firm to exceed 100 trillion won in operating profit, a feat largely driven by skyrocketing demand for AI-related semiconductor products.
A Chinese developer has revoked public access to the agentic AI pentesting tool ARTEX after it was linked to cyberattacks on multiple South Korean banks. CrowdStrike identified a suspect who combined ARTEX with several LLMs to compromise financial institutions and inadvertently expose his identity through an AI-generated resume.
Drone attacks targeted and disabled two of Yandex's six primary data hubs in Russia, including a major facility in Ryazan Oblast that housed nearly 2,700 Nvidia A100 GPUs. The attacks caused widespread service interruptions and exposed structural vulnerabilities in Russia's cloud infrastructure.
Scientists have developed SPERA, a joint-embedding predictive architecture designed to serve as a general-purpose foundation model for electroencephalography (EEG) data. Pretrained on 80,000 hours of EEG data, it demonstrates high parameter efficiency and robust performance across clinical, cognitive, and BCI applications.
Researchers have introduced a multimodal multitask framework combining MRI encoders, clinical Transformers, and ODE-GRU dynamics to improve Alzheimer's disease diagnosis and progression prediction. The approach addresses irregular visits and missing modalities, achieving high predictive accuracy across major datasets like ADNI and OASIS.
Researchers demonstrated that fine-tuned Qwen3.5 small language models (0.8B) outperform prompted frontier models like GPT-5 in grammar concept annotation while reducing serving costs by 16x. The approach successfully improved learner engagement and business metrics in an online language learning platform.
Cryptographer Matthew Green has expressed concerns that AI could introduce severe vulnerabilities to existing public-key encryption algorithms faster than human standards can adapt. This highlights a critical gap between the speed of AI-driven surprises and the slow pace of security standard replacement.
Project Strata rewrote its Git history to strip out references to Claude in commit messages, drawing criticism from the community regarding transparency around AI-assisted development.
Researchers proposed Diffu-LoRA, a parameter-efficient method that optimizes layer-wise rank allocation for personalized text-to-image diffusion models. This approach improves subject fidelity and prompt alignment while keeping the pretrained backbone frozen.
Developer Simon Willison has released ttok 1.0, updating the tool to default to GPT-5 and GPT-6 tokenizers based on findings that recent OpenAI models share identical tokenization characteristics.
Is the future of frontier AI open or closed? ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo and Elad Gil to discuss the launch of Beam, the company’s 500 billion parameter open-weight…
Nathan Labenz reports back from The Curve conference, sharing off-the-record insights from frontier lab leaders on shortened timelines and whether humanity is approaching an intelligence threshold it…
OutSystems CEO Woodson Martin joins Nathan to discuss how enterprise software can evolve at AI speed without breaking critical operations. Martin explains how an intermediate layer of abstract…
In this episode, Diogo Almeida, co-founder and CEO of TypeSafe, joins us to discuss Jev, TypeSafe’s recently released model for bringing fast, reliable intelligence directly into software. We explore…
Yash Patil, a former OpenAI researcher who worked on Codex and now runs Applied Compute, argues that post-training is the most overlooked part of the AI stack and the key to owning your intelligence.…
At 100,000-accelerator scale, something fails multiple times an hour, which is why Google's Chief Technologist for AI Infrastructure Amin Vahdat thinks FLOPS is a vanity metric. The metric that…
Keerthana Gopalakrishnan, Research Lead for Gemini Robotics at Google DeepMind, returns to discuss the release of Gemini Robotics 2, whole-body intelligence, and the pursuit of generalist humanoid…