AI Research

Twitter

Facebook

Copy Link

Key Facts

FocusScaling, Reasoning, Safety, Multimodality

Key OrgsGoogle DeepMind, Stanford HAI, MILA, MIT CSAIL, Allen AI

Key EventsNeurIPS, ICML, ICLR

Top News

Physical Intelligence seeks to raise $1 billion, doubling valuation to $11 billion in four months

Physical Intelligence, a two-year-old robotics startup founded by former Google DeepMind researchers, is in discussions to raise approximately $1 billion in new funding at a valuation exceeding $11 billion. The deal would double the company's $5.6 billion valuation from just four months ago, with Founders Fund and Lightspeed Venture Partners among the investors as the AI robotics sector attracts growing capital.

Startups3 sources

Physical Intelligence seeks to raise $1 billion, doubling valuation to $11 billion in four months

Startups3 sources

NeurIPS reverses controversial ban after China threatens boycott over US sanctions policy

The world's premier AI research conference NeurIPS sparked international controversy after announcing restrictions on US-sanctioned entities like Huawei and SMIC. China's largest science federation CAST responded with a boycott, redirecting funding to domestic conferences and refusing to recognize NeurIPS publications. The organizers quickly reversed course, calling it a miscommunication, but the incident highlights deepening geopolitical tensions threatening global scientific collaboration in AI research.

Policy5 sources

NeurIPS reverses controversial ban after China threatens boycott over US sanctions policy

Policy5 sources

GPT Models Rate Literary Nonsense Highly, Raising Concerns About AI Reasoning Biases

A German researcher discovered that OpenAI's GPT models consistently rate pseudo-literary nonsense higher than simple coherent text, even with reasoning features activated. The findings reveal critical AI reasoning biases that could affect AI development implications, especially as AI models increasingly evaluate each other's work with minimal human oversight.

Technology2 sources

GPT Models Rate Literary Nonsense Highly, Raising Concerns About AI Reasoning Biases

Technology2 sources

Google's TurboQuant slashes AI memory usage by 6x while maintaining quality and speed

Google Research introduced TurboQuant, a compression algorithm that reduces large language model memory requirements by at least 6x without sacrificing accuracy. The breakthrough targets the key-value cache bottleneck, delivering up to 8x faster performance on Nvidia H100 GPUs while compressing data to just 3 bits. But experts warn it won't solve the broader memory shortage.

Technology14 sources

Google's TurboQuant slashes AI memory usage by 6x while maintaining quality and speed

Technology14 sources

MIT's ultrasound wristband lets wearers control robotic hands with their own hand movements

MIT engineers have created a wearable ultrasound wristband that captures the intricate dance of 34 muscles, 27 joints, and over 100 tendons in the human hand. Using AI to decode ultrasound images of wrist muscles and tendons, the device enables wearers to control robotic hands in real-time—playing piano, shooting hoops, or manipulating virtual objects with unprecedented precision.

Science5 sources

MIT's ultrasound wristband lets wearers control robotic hands with their own hand movements

Science5 sources

Adversarial AI uncovers hidden brain mechanisms and potential treatments for disorders of consciousness

Researchers at UCLA developed a generative adversarial artificial intelligence framework that pits two AI models against each other to simulate conscious and unconscious brain states. The system, trained on over 680,000 neuroelectrophysiology samples, identified previously unknown mechanisms behind coma and predicted that stimulation of the subthalamic nucleus could help restore consciousness in patients with brain injuries.

Science2 sources

Adversarial AI uncovers hidden brain mechanisms and potential treatments for disorders of consciousness

Science2 sources

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.

Science3 sources

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

Science3 sources

AI could boost euro area productivity growth by 4% over decade, but Europe lags behind U.S.

The European Central Bank projects AI could lift euro zone productivity growth by over 4 percentage points in 10 years if adoption accelerates. But ECB chief economist Philip Lane warns that prolonged energy shocks and Europe's dependence on imported technology—spending nearly 250 billion euros annually on foreign royalties—could slow progress as the bloc trails the U.S. in AI innovation.

Business2 sources

AI could boost euro area productivity growth by 4% over decade, but Europe lags behind U.S.

Business2 sources

Autoscience raises $14 million to build AI models using AI, signaling automation of research itself

Autoscience has emerged with $14 million in seed funding to develop an autonomous AI research lab where nonhuman AI scientists create specialized machine learning models. The startup's AI agent Carl already produced a peer-reviewed research paper with minimal human involvement, accepted to ICLR 2025. This development suggests AI research itself may soon be automated, potentially disrupting even AI engineers' roles.

Startups2 sources

Autoscience raises $14 million to build AI models using AI, signaling automation of research itself

Startups2 sources

Xiaomi commits $8.7 billion to AI as MiMo-V2-Pro model drives stock surge and investor optimism

Xiaomi announced a massive $8.7 billion AI investment over the next three years, following the successful launch of its MiMo-V2-Pro large language model. The Chinese tech giant's stock jumped, making it the top performer on the Hang Seng Tech Index as CEO Lei Jun outlined ambitious plans to compete in the rapidly evolving AI agent market.

Technology2 sources

Xiaomi commits $8.7 billion to AI as MiMo-V2-Pro model drives stock surge and investor optimism

Technology2 sources

Google DeepMind hires Bridgewater's Jasjeet Sekhon as chief strategy officer for AGI push

Jasjeet Sekhon, former chief scientist at Bridgewater Associates, is joining Google DeepMind as chief strategy officer. The move strengthens Google's AI strategy as it competes with OpenAI and Anthropic. Sekhon will work directly with DeepMind founder Demis Hassabis to develop artificial general intelligence safely while continuing to serve on Bridgewater's board.

Business5 sources

Google DeepMind hires Bridgewater's Jasjeet Sekhon as chief strategy officer for AGI push

Business5 sources

Fitbit AI Coach will read your medical records next month as Google pushes deeper into healthcare

Google announced that Fitbit users can soon connect medical records to its AI-powered health coach for personalized guidance. The update includes lab results, medications, and visit history integration, alongside a 15% improvement in sleep tracking accuracy. While Google promises data privacy and says health information won't be used for ads, concerns remain about AI hallucinations and over-reliance on chatbots for medical advice.

Technology5 sources

Fitbit AI Coach will read your medical records next month as Google pushes deeper into healthcare

Technology5 sources

Apple's LiTo AI model reconstructs 3D objects with realistic lighting from a single image

Apple researchers have developed LiTo, an AI model that reconstructs 3D objects from a single image while preserving realistic lighting effects like reflections and highlights across different viewing angles. The Surface Light Field Tokenization approach uses latent space to jointly model object geometry and view-dependent appearance, outperforming existing methods that typically require multiple images.

Technology3 sources

Apple's LiTo AI model reconstructs 3D objects with realistic lighting from a single image

Technology3 sources

Humanoid robot masters tennis using AI that learns from imperfect motion data

Chinese AI robotics company Galbot has taught a Unitree G1 humanoid robot to play tennis against humans with remarkable success. Using a new system called LATENT, researchers trained the robot with just five hours of motion fragments rather than complete tennis sequences, achieving up to 96% success rates in simulation. The breakthrough suggests robots can learn complex physical tasks from imperfect data.

Technology3 sources

Humanoid robot masters tennis using AI that learns from imperfect motion data

Technology3 sources

UK government invests £45 million in AI supercomputer to accelerate fusion research

The UK is investing £45 million in Sunrise, an AI supercomputer dedicated to fusion research at the UK Atomic Energy Authority's Culham campus. Set to launch in June, the 1.4MW system will model plasma behavior and test nuclear fusion reactor designs virtually before costly physical experiments.

Technology2 sources

UK government invests £45 million in AI supercomputer to accelerate fusion research

Technology2 sources

Prev 1…111213…31 Next