News
Google DeepMind Launches D4RT AI Model for Real-Time 4D Reconstruction
3+ hour, 14+ min ago (994+ words) Google DeepMind has released D4RT, a unified AI model for 4D scene reconstruction that runs 18 to 300 times faster than comparable methods, targeting robotics and AR applications. Google DeepMind released D4RT on Friday, a model for 4D scene reconstruction that combines depth estimation, spatiotemporal correspondence,…...
Google Vids Adds TikTok-Style Animated Captions
5+ hour, 47+ min ago (392+ words) Google has introduced stylized animated captions for Vids, offering TikTok-style text effects across all Workspace tiers with manual editing for accuracy. Google has unveiled animated, stylized captions for Google Vids, letting Workspace users add TikTok-style text effects without third-party editors....
GPT-5.2 Pro Solves Decades-old Math Problem, but Experts Say It Reveals AI's Limits as Much as Its Potential
5+ day, 12+ hour ago (1290+ words) OpenAI's GPT-5.2 Pro has solved multiple decades-old Erd's math problems, but Fields Medalist Terence Tao says the wins demonstrate speed over insight. When OpenAI's ChatGPT Pro solved a 50-year-old mathematical conjecture last week, headlines celebrated a breakthrough. Fields Medalist Terence…...
DeepSeek has introduced Manifold-Constrained Hyper-Connections (mHC), a novel architecture that stabilizes AI training and enables scaling on constrained hardware.
3+ week, 1+ day ago (895+ words) Detailed in the official technical paper released Tuesday, the method restores signal integrity in deep networks by projecting connections onto a mathematical manifold. This fix reportedly eliminates training instability while adding less than 7% to total compute time. Such efficiency is…...
Microsoft in November launched a managed memory feature for its Foundry Agent Service, enabling enterprise AI agents to retain long-term user context without custom databases.
3+ week, 4+ day ago (719+ words) Microsoft in November launched a managed memory feature for its Foundry Agent Service, enabling enterprise AI agents to retain long-term user context without custom databases. Microsoft is attempting to solve the "goldfish memory" problem plaguing enterprise AI by integrating long-term…...
Gemma Scope 2: New Google Tools Let Researchers Trace AI 'Thought' Circuits
1+ mon, 5+ day ago (1446+ words) Gemma Scope 2,is a comprehensive open-source suite trained on 110 petabytes of data to map internal reasoning circuits across the entire Gemma 3 model family. Opening the "black box" of artificial intelligence to new levels of scrutiny, Google DeepMind has released Gemma…...
Alpamayo-R1: NVIDIA Releases Vision Reasoning Model and Massive 1,727-Hour Dataset for Autonomous Driving
1+ mon, 3+ week ago (1205+ words) NVIDIA has unveiled Alpamayo-R1, a reasoning-based AV model, and a 1,727-hour open dataset to facitlitate autonomous vehicle development. NVIDIA is attempting to solve the "black box" problem of self-driving cars by open-sourcing the cognitive architecture behind them. At the NeurIPS…...
Study Shows How AI Agents Break Online Surveys and Polling, Posing Existential Threat to Data Science
2+ mon, 9+ hour ago (1287+ words) Autonomous AI agents can now pass 99.8% of survey attention checks, threatening the integrity of scientific and political data. Online polling and social science research face an existential crisis after a new study revealed that autonomous AI agents can now reason…...
Harvard's popEVE AI Model Identifies 123 Novel Disease Genes
2+ mon, 10+ hour ago (972+ words) Harvard Medical School has unveiled popEVE, an genetic disease AI model that outperforms Google DeepMind's AlphaMissense by reducing false positive genetic diagnoses. Challenging the dominance of Google DeepMind in computational biology, researchers at Harvard Medical School have unveiled popEVE, a…...
Anthropic Releases Open-Source AI Bias Test, Pitting Claude Against GPT-5, Gemini, Grok in Race for AI Neutrality
2+ mon, 1+ week ago (927+ words) Anthropic has released an open-source framework to measure political bias in AI, claiming its Claude model is more even-handed than OpenAI's GPT-5 and Meta's Llama 4. Anthropic on Thursday released an open-source framework to measure political "evenhandedness" in AI models, positioning…...