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

DeepSeek unveils mHC architecture that could reshape how developers train advanced AI models

Chinese AI firm DeepSeek has released research on Manifold-Constrained Hyper-Connections (mHC), a new AI training method that addresses signal degradation in neural networks. The framework could enable smaller developers to build frontier models without massive computational costs, potentially previewing the architecture behind the anticipated DeepSeek R2 model expected around February's Spring Festival.

Science7 sources

DeepSeek unveils mHC architecture that could reshape how developers train advanced AI models

Science7 sources

OpenAI employee compensation reaches $1.5 million per worker, shattering Silicon Valley records

OpenAI is paying its 4,000 employees an average of $1.5 million each in stock-based compensation, far exceeding what Google, Facebook, or any major tech company offered before going public. The aggressive pay structure reflects the intense competition for AI talent, with costs projected to reach 46% of revenue by 2025—the highest among major tech startups analyzed over the past 25 years.

Business3 sources

OpenAI employee compensation reaches $1.5 million per worker, shattering Silicon Valley records

Business3 sources

Large language models achieve under 1% accuracy at basic multiplication, new study reveals

Despite handling complex coding tasks, large language models fail at four-digit multiplication, achieving less than 1% accuracy with standard training. Researchers from University of Chicago, MIT, Harvard, and Google DeepMind discovered the culprit: models can't store and retrieve intermediate computations. But a specialized Implicit Chain of Thought method achieved 100% accuracy by teaching models to internalize reasoning processes.

Science2 sources

Large language models achieve under 1% accuracy at basic multiplication, new study reveals

Science2 sources

Silicon Valley summit reveals China's dominance in humanoid robots as AI sparks new momentum

Over 2,000 robotics engineers and investors gathered at Silicon Valley's Humanoids Summit to showcase the latest in humanoid robots powered by artificial intelligence. While the boom in artificial intelligence has revived interest in the sector, China emerged as the clear leader with about 20 companies raising at least $100 million each. Despite growing investor interest driven by generative AI breakthroughs, skepticism remains high about when these robots will achieve widespread adoption in workplaces and homes.

Technology3 sources

Silicon Valley summit reveals China's dominance in humanoid robots as AI sparks new momentum

Technology3 sources

Consciousness May Require Physical Brain Computation, Not Just Abstract Code

A new theoretical framework proposes biological computationalism as a third path between computational functionalism and biological naturalism. Researchers argue that consciousness cannot be reduced to abstract information processing because brain computation is inseparable from its physical, hybrid, and energy-constrained dynamics. This challenges assumptions about whether digital AI can truly recreate conscious experience.

Science2 sources

Consciousness May Require Physical Brain Computation, Not Just Abstract Code

Science2 sources

AI Tool Predicts Underfeeding Risk in ICU Patients on Ventilators, Enabling Early Intervention

Researchers at Mount Sinai developed NutriSighT, an AI tool that predicts which critically ill ICU patients on ventilators face underfeeding risk during their first week of care. The study, published in Nature Communications, found that 41-53% of patients were underfed by day three, with the system updating predictions every four hours to help clinicians intervene earlier.

Health2 sources

AI Tool Predicts Underfeeding Risk in ICU Patients on Ventilators, Enabling Early Intervention

Health2 sources

Yann LeCun vs Demis Hassabis: Public Clash Over General Intelligence Reopens AGI Debate

Meta's Chief AI Scientist Yann LeCun and Google DeepMind CEO Demis Hassabis engaged in a heated public disagreement over whether general intelligence exists. LeCun argues human intelligence is highly specialized, while Hassabis defends it as genuinely general. The debate carries significant implications for artificial general intelligence development across the AI industry.

Technology3 sources

Yann LeCun vs Demis Hassabis: Public Clash Over General Intelligence Reopens AGI Debate

Technology3 sources

New Open-Source Tool RNACOREX Maps Gene Regulation Networks to Predict Cancer Survival

Researchers at the University of Navarra in Spain have developed RNACOREX, an open-source software platform that identifies gene regulation networks linked to cancer survival. The tool analyzes thousands of biological molecules simultaneously to detect molecular interactions often missed by traditional methods, providing an explainable alternative to AI models while predicting patient survival with comparable accuracy.

Science2 sources

New Open-Source Tool RNACOREX Maps Gene Regulation Networks to Predict Cancer Survival

Science2 sources

AI image generators converge on just 12 visual clichés, study reveals creative limitations

A study published in Patterns journal reveals AI image generators repeatedly default to the same 12 generic visual motifs when generating images autonomously. Researchers paired AI models in a visual telephone experiment, finding that despite diverse starting prompts, systems like Stable Diffusion XL consistently produced Eurocentric themes like Gothic cathedrals and Parisian nightscapes—what they call 'visual elevator music.'

Science3 sources

AI image generators converge on just 12 visual clichés, study reveals creative limitations

Science3 sources

Meta develops Mango AI model for images and videos, targets first half of 2026 launch

Meta is building Mango, a new image and video AI model, alongside Avocado, a text-based large language model, both slated for early 2026 release. Chief AI officer Alexandr Wang revealed the roadmap during an internal Q&A, emphasizing improved coding capabilities and world models. The move intensifies competition with OpenAI, Google, and Anthropic in the generative AI race.

Technology4 sources

Meta develops Mango AI model for images and videos, targets first half of 2026 launch

Technology4 sources

Yann LeCun confirms Advanced Machine Intelligence startup, targets €3 billion valuation pre-launch

Renowned AI scientist Yann LeCun confirmed his new startup Advanced Machine Intelligence, seeking €500 million at a €3 billion valuation before launch. The venture will develop world model AI systems that understand the physical world, with Alexandre LeBrun from Nabla as CEO. This marks one of the largest pre-launch AI fundraising rounds.

Startups7 sources

Yann LeCun confirms Advanced Machine Intelligence startup, targets €3 billion valuation pre-launch

Startups7 sources

Skana Robotics enables underwater robots to communicate securely using AI algorithms

Tel Aviv-based Skana Robotics has developed an AI-powered capability that allows underwater robots to communicate across long distances without surfacing. The breakthrough addresses a critical challenge in defense operations and infrastructure protection, where submersibles traditionally had to expose themselves by rising to the surface to transmit data.

Technology2 sources

Skana Robotics enables underwater robots to communicate securely using AI algorithms

Technology2 sources

Patronus AI unveils Generative Simulators to fix 63% failure rate plaguing AI agents

Patronus AI introduced Generative Simulators, a training architecture that creates adaptive simulation environments to address the 63% failure rate AI agents face on complex tasks. The technology dynamically generates new challenges and provides continuous feedback, moving away from static benchmarks that fail to predict real-world performance.

Technology2 sources

Patronus AI unveils Generative Simulators to fix 63% failure rate plaguing AI agents

Technology2 sources

Databricks raises $4B at $134B valuation as AI business drives 55% revenue growth

Databricks has closed a $4 billion Series L funding round at a $134 billion valuation, just three months after hitting $100 billion. The data intelligence company now generates $4.8 billion in annual revenue, with over $1 billion from AI products. The funding will accelerate development of Lakebase, Agent Bricks, and Databricks Apps while supporting acquisitions and global expansion.

Business6 sources

Databricks raises $4B at $134B valuation as AI business drives 55% revenue growth

Business6 sources

OpenAI-backed Chai Discovery raises $130M at $1.3B valuation as AI transforms drug development

Chai Discovery, an OpenAI-backed biotech startup, secured $130 million in Series B funding at a $1.3 billion valuation. Led by General Catalyst and Oak HC/FT, the round brings total funding to over $225 million. The company's Chai 2 AI model designs custom antibodies from scratch, solving complex drug development challenges that previously took years in just weeks.

Startups4 sources

OpenAI-backed Chai Discovery raises $130M at $1.3B valuation as AI transforms drug development

Startups4 sources

Prev 1…222324…33 Next