Brief

Researchers have developed a novel approach called TIME (Temporally Informed Motion Embedding) that leverages motion for efficient video representation learning. This method uses a masked autoencoder trained on synthetic motion data, specifically point-tracks, to reconstruct missing movements. By focusing on motion, TIME significantly reduces the need for massive training datasets and bypasses language-dependent paradigms, leading to better temporal understanding and fine-grained concept learning. AI

Researchers have introduced Toto 2.0, a suite of five open-weight time series forecasting models that demonstrate the effectiveness of scaling foundation models. The models, trained using a single recipe, show improved forecast quality as their parameter count increases from 4 million to 2.5 billion. Toto 2.0 has achieved state-of-the-art results on three key forecasting benchmarks: BOOM, GIFT-Eval, and TIME. AI

Amazon has launched a new feature for its Alexa Plus AI assistant that allows users to generate custom podcast episodes on demand. Users can specify a topic, and Alexa will research and create an episode with AI-generated hosts, offering options to adjust length and focus. This expansion moves Alexa beyond a simple voice assistant into a personalized AI content creator, drawing information from over 200 news partners to enhance accuracy. AI

Amazon has begun producing AI-generated podcasts by repurposing content from its Alexa assistant. These podcasts feature two AI-generated hosts discussing various topics, including cultural trends and the music industry. The company claims journalists will fact-check the content, partnering with major news organizations to ensure accuracy, though critics argue this is an attempt to generate engagement slop with minimal human oversight. AI