The Batch | DeepLearning.AI (Page 42)

Robots filming and recording in a scenic grassy field under a sunny, rural landscape.

Data Points

Cobot’s Proxie robot tackles warehouse tasks: VBench++, a new benchmark suite for AI video

Anthropic and Amazon strengthen their ties. Windsurf blends copilots with agents in one IDE. Mistral introduces Pixtral Large to its APIs and chat platform. H’s first product launch is a business agent.

Robot walking through a traditional wooden library with high shelves and warm lighting.

Data Points

DeepSeek’s R1 seeks to match OpenAI’s o1: Microsoft will train future model on big publisher’s books

Flux’s new image tools compete with Adobe and other photo apps. Qwen speeds up its 2.5 model while boosting context window. Rabbit launches Teach Mode in beta for its R1 devices. Banks embrace third-party AI, but want to control their data.

Two people reading in bed, one with a book on library functions and a head labeled with AI layers.

Letters

AI Is Part of Your Online Audience: Some webpages are written not for humans but for large language models to read. Developers can benefit by keeping the LLM audience in mind.

A small number of people are posting text online that’s intended for direct consumption not by humans, but by LLMs (large language models).

The Batch Newsletter

Next-Gen Models Show Limited Gains, Real-Time Video Generation, China AI Chips Blocked, Transformer Training Streamlined

The Batch AI News and Insights: A small number of people are posting text online that’s intended for direct consumption not by humans, but by LLMs (large language models).

Efficient Foundations animation showing layered AI model components.

Machine Learning Research

More-Efficient Training for Transformers: Researchers reduce transformer training costs by 20% with minimal performance loss

Researchers cut the processing required to train transformers by around 20 percent with only a slight degradation in performance.

Hardware

Further Chip Restrictions on China: TSMC stops advanced chip production for China on U.S. orders

The largest manufacturer of AI chips told its Chinese customers it would stop fabricating their most advanced designs, further limiting China’s access to AI hardware.

Comparison of Minecraft terrain with and without player modifications.

Machine Learning Research

No Game Engine Required: AI creates an interactive Minecraft-like world in real time

A real-time video generator lets you explore an open-ended, interactive virtual world — a video game without a game engine.

Graph showing test loss decreases with more tokens and larger model sizes (103-109 parameters).

Machine Learning Research

Next-Gen Models Show Limited Gains: AI giants rethink model training strategy as scaling laws break down

Builders of large AI models have relied on the idea that bigger neural networks trained on more data and given more processing power would show steady improvements. Recent developments are challenging that idea.

A group of highly advanced robots dressed in traditional firefighter uniforms in a rugged forest

Data Points

Finding the limits of pretraining and quantization: New EU draft guidelines for regulating top models

Google AI detects wildfires from satellite images. Baidu shows off image RAG, no-code tools, and smart glasses. ChatGPT apps link up with desktop coding tools. New LoRA tools improve image models’ consistency and text generation.

A court scene with a humanoid robot judge presiding over a courtroom.

Data Points

OpenAI and Google’s fine-tuning disappoints: Alibaba’s latest Qwen open source code model matches GPT-4

MLPerf tests show Nvidia’s Blackwell chips excel at training models. Google releases AlphaFold 3 code and parameters (with restrictions). DeepSeek’s two-way image model gets even better. Judge tosses copyright lawsuit against OpenAI.

Man with tools says, “I optimized for tool use!” Woman at computer replies, “Should’ve optimized for computer use!”

Letters

From Optimizing for People to Optimizing for Machines: Why large language models are increasingly fine-tuned to fit into agentic workflows

Large language models (LLMs) are typically optimized to answer peoples’ questions.

The Batch Newsletter

Llama On the Battlefield, Mixture of Experts Pulls Ahead, Open Agentic Platform, Voter Support Chatbot

The Batch AI News and Insights: Large language models (LLMs) are typically optimized to answer peoples’ questions.

Latest

Cobot’s Proxie robot tackles warehouse tasks: VBench++, a new benchmark suite for AI video

DeepSeek’s R1 seeks to match OpenAI’s o1: Microsoft will train future model on big publisher’s books

AI Is Part of Your Online Audience: Some webpages are written not for humans but for large language models to read. Developers can benefit by keeping the LLM audience in mind.

Next-Gen Models Show Limited Gains, Real-Time Video Generation, China AI Chips Blocked, Transformer Training Streamlined

More-Efficient Training for Transformers: Researchers reduce transformer training costs by 20% with minimal performance loss

Further Chip Restrictions on China: TSMC stops advanced chip production for China on U.S. orders

No Game Engine Required: AI creates an interactive Minecraft-like world in real time

Next-Gen Models Show Limited Gains: AI giants rethink model training strategy as scaling laws break down

Finding the limits of pretraining and quantization: New EU draft guidelines for regulating top models

OpenAI and Google’s fine-tuning disappoints: Alibaba’s latest Qwen open source code model matches GPT-4

From Optimizing for People to Optimizing for Machines: Why large language models are increasingly fine-tuned to fit into agentic workflows

Llama On the Battlefield, Mixture of Experts Pulls Ahead, Open Agentic Platform, Voter Support Chatbot