Analytics DeepLearning.AI - The Batch | DeepLearning.AI (Page 3)

Cartoon robots with square heads and antennae sit in rows on an assembly line, each smiling while assembling gears, boxes, and tools.

Letters

Agents Running in Parallel Get There Faster: Parallel agents can accelerate AI systems as test-time compute scales up.

Parallel agents are emerging as an important new direction for scaling up AI. AI capabilities have scaled with more training data, training-time compute, and test-time compute.

Table comparing DINO, DINOv2, DINOv3, SigLIP 2, and PE on segmentation, depth estimation, tracking, and classification tasks.

Machine Learning Research

Better Image Processing Through Self-Supervised Learning: Meta’s DINOv3 gets an updated loss term and improved vision performance

DINOv2 showed that a vision transformer pretrained on unlabeled images could produce embeddings that are useful for a wide variety of tasks. Now it has been updated to improve the performance of its embeddings in segmentation and other vision tasks.

Researcher adjusts sensors on a quadruped robot disguised as a Tibetan antelope in a high-altitude reserve.

Robotics

Robot Antelope Joins Herd: Chinese scientists disguise modified robot dog as antelope to study herd behavior

Researchers in China disguised a quadruped robot as a Tibetan antelope to help study the animals close-up.

Infographic showing Mistral Large 2’s environmental impact across emissions, water, and material use.

Science

Mistral Measures LLM Consumption of Energy, Water, and Materials: French AI startup discloses full lifecycle consumption and emissions for Mistral Large 2

The French AI company Mistral measured the environmental impacts of its flagship large language model.

Hand holding a Pixel 10 smartphone using Google’s Magic Cue AI assistant, which suggests a location reply in a text message conversation.

Hardware

Proactive AI Assistance for Phones: Inside Magic Cue, Google’s new AI assistant for Pixel 10

Google’s latest smartphone sports an AI assistant that anticipates the user’s needs and presents helpful information without prompting.

Modern office team collaborating on AI data analytics with dashboards and charts on multiple computer screens.

Data Points

Perplexity offers AI news subscription plan: Nvidia Nano 2 employs Mamba for speed

Cohere’s new reasoning mode for open-weights Command A. Hacking MCP to turn Claude into an image generator. Google’s new AI deal with the U.S. government. Meta’s partnership with image generator Midjourney.

Scientists in lab coats analyzing black hole simulation on high-tech monitors in a futuristic control room.

Data Points

DeepSeek releases a hybrid reasoning model: OpenAI unveils a new subscription plan for India

Alibaba’s new text and image editing features. Adobe’s tool to chat with your PDFs and other docs. Browsemaster, a new framework for agentic search. Surya, an IBM/NASA model that predicts solar weather.

Andrew Ng speaks at the August 2025 Buildathon hosted by AI Fund and DeepLearning.AI. A packed audience watches the event, and groups of participants collaborate on laptops.

Letters

How Non-Coders Built 5 Software Products in 6½ Hours: At Buildathon on August 16, coders and non-coders alike showed how much AI is changing software development.

On Saturday at the Buildathon [http://buildathon.ai] hosted by AI Fund and DeepLearning.AI, over 100 developers competed to build software products quickly using AI assisted coding.

The Batch Newsletter

China Questions Nvidia, When Models Memorize, Mixture of Video Experts, OpenAI & Oracle Join Forces

The Batch AI News and Insights: On Saturday at the Buildathon [http://buildathon.ai] hosted by AI Fund and DeepLearning.AI, over 100 developers competed to build software products quickly using AI assisted coding.

Charts showing how much information language models memorize based on model size and training dataset size; one graph uses random data, the other uses real-world text.

Machine Learning Research

Does Your Model Generalize or Memorize: Researchers find models with more parameters copy more bits from training sets

Benchmarks can measure how well large language models apply what they’ve learned from their training data to new data, but it’s harder to measure the degree to which they simply memorized their training data. New work proposes a way to gauge memorization.

Aerial view of a large, partially constructed data center surrounded by parked vehicles and red soil in Abilene, Texas

Business

OpenAI Turns to Oracle for Compute: A new $30 billion, 4.5 gigawatt data center offshoot of the Stargate Project

OpenAI is working with Oracle to build its next chunk of processing power, a $30 billion outgrowth of the partners’ $500 billion Stargate project and a sign of OpenAI’s ongoing thirst for computation.

Three AI-generated video clips: a man vaulting over a moving car, a gymnast flipping on a plane wing, and a rabbit ice skating in pink boots.

Machine Learning Research

Mixture of Video Experts: Alibaba’s Wan 2.2 video models adopt a new architecture to sort noisy from less-noisy inputs

The mixture-of-experts approach that has boosted the performance of large language models may do the same for video generation.