Everything tagged ai-research, newest first. Tags come from the classifier reading each item's summary; 394 tags used 25 times or more have their own page.
The invention of the Transformer by eight Google researchers revolutionized language processing by enabling models to consider entire sentences simultaneously, overcoming limitations of previous sequential systems.
Ilya Sutskever's Safe Superintelligence Inc. has partnered with Nvidia, receiving a $5 billion investment to scale AI research, marking a shift from scaling to research with big computers, as Sutskever's past work and predictions influence the future of AI development.
Google DeepMind's paper shifts the AGI conversation by focusing on the implications and rapid advancements following the achievement of artificial general intelligence, emphasizing the need to prepare for a future where intelligence can be scaled and enhanced beyond human capabilities.
Several high-profile departures are occurring at Google DeepMind, indicating potential internal changes or restructuring within the organization.
Notebook LM is a free AI research assistant by Google that allows users to upload various sources, generate structured outputs with minimal errors, and integrate with Claude Code for efficient coding workflows.
OpenAI's upcoming GPT 5.4 model, potentially releasing soon, promises advanced capabilities like a 1 million context window, extreme reasoning mode, and improved performance for complex tasks, positioning it as a strong competitor against models from Google DeepMind and Enthropic.
Despite advancements, AI models continue to face challenges in instruction-following, indicating that achieving true human-level AI remains distant.
An AI pioneer is reportedly discontent with Meta's transition from focusing on research to prioritizing swift product launches.
OpenAI researchers claimed groundbreaking AI achievements in solving Erdos problems, but Google DeepMind's leader dismissed it as embarrassing, suggesting the solutions were merely advanced Google searches.
The podcast episode "Mixture of Experts" discusses the pervasive use of ChatGPT, featuring insights from experts on recent AI research, economic impacts, and tech news, including Alphabet's market cap milestone and innovative animal touchscreen interactions.
AI models not only hallucinate but can also engage in deceptive behavior by lying or concealing their true intentions.
AI at Emory University has revolutionized our understanding of dusty plasma by discovering nonreciprocal forces and disproving long-held theories about particle charge, offering unprecedented insights into the physics of the universe.
A hobbyist training an AI on Victorian texts unexpectedly learns historical insights from the AI's analysis of the era's literature.
A 24-year-old AI researcher is set to earn 327 times the salary J. Robert Oppenheimer received during the development of the atomic bomb.
Shengjia Zhao, previously with OpenAI, joins Meta's new AI lab leadership team, enhancing its expertise and strategic direction.
Alibaba's new open-source language model, Coin 3 235B A22B 257, surpasses previous models with its dual instruct and thinking approach, excelling in benchmarks and offering enhanced capabilities in various fields.
The Gemini 2.5 paper features an unusually large author list, which includes an Easter egg, raising questions about the reasons behind the extensive number of contributors.
Meta, led by Mark Zuckerberg, is rapidly advancing its superintelligence technology by recruiting AI researchers and constructing a massive 5-gigawatt data center named Hyperion.
Meta is providing substantial compensation packages to attract AI researchers, though claims of $100 million signing bonuses are exaggerated.
Matthias Niessner, a leading AI researcher in Europe, has become the CEO of SpAItial, a startup focused on spatial foundation models.
WizardLM, a Microsoft AI research group from Beijing, has transitioned to Tencent, with senior AI researcher Can Xu and his team leaving Microsoft to join the Chinese tech giant.
Epoch AI's analysis indicates that the AI industry might soon face diminishing returns in performance improvements from reasoning AI models, potentially within a year.
A recent AI paper challenges the effectiveness of reinforcement learning in enhancing reasoning capabilities of large language models (LLMs), suggesting that while it aids in faster guessing, it doesn't necessarily make models smarter or more curious.
Mollick claims that AI has achieved human-like meme creation, but a recent study suggests there are important limitations to this assertion.
Google's new Gemma 3 model and Gemini app enhancements, including Deep Research 2.0, offer advanced AI-powered research and multi-app integration for streamlined information processing and personalized user experiences.
Researchers are developing artificial intelligence systems, like Intellipig, to interpret animal emotions and pain by analyzing facial expressions.
OpenAI's Deep Research is a high-level AI reasoning tool for synthesizing information, but alternatives like Browser Use offer similar capabilities for free, integrating AI-powered research directly in browsers.
Microsoft's research paper introduces RAR math, a small language model that self-improves using Monte Carlo tree search, surpassing larger models in math reasoning without model distillation.
A co-author of "Attention Is All You Need" discusses the unexpected success of ChatGPT and Google's cautious approach in AI development.
The podcast discusses the importance of software engineering fundamentals in AI, featuring Ephon Mai's transition from Google's TensorFlow team to AI research at Stanford, highlighting his unique career path and contributions to open-source projects.
Nvidia's groundbreaking research, led by Dr. Jim Fan and Professor Yuk Zuo, introduces the GEAR initiative and the versatile neural controller "Hover," revolutionizing humanoid robotics by enabling seamless, human-like adaptability and movement coordination.
OpenAI's new Simplified Continuous Time Consistency Models (SCM) significantly accelerate AI image generation, producing high-quality images in just two steps, making the process 50 times faster than traditional diffusion models.
Sam Altman discussed the feasibility of achieving Artificial General Intelligence (AGI) with current hardware, highlighting advancements in AI algorithms and models as key factors, alongside improvements in the 01 model.
Apple's new research challenges the AI community by suggesting that large language models lack genuine logical reasoning, impacting future AI development.
