Analytics DeepLearning.AI - The Batch

Film director with megaphone on busy movie set surrounded by crew, bank facade, lighting, and equipment.

Data Points

Sora no more; OpenAI shuts down video maker: Tencent introduces ClawBot, an OpenClaw wrapper for WeChat

Anthropic’s multi-agent harness design. A new voice agent benchmark. Arm’s chip design for data center CPU. U.S. White House’s AI proposal to Congress.

Cozy home office with man on couch using smartphone, dual monitors displaying code.

Data Points

MiniMax’s M2.7 model competes with Gemini and Opus: Cursor’s Composer 2 does the same, but for code

Microsoft’s latest image model. Google’s Stitch, a platform for designers. Claude Code Channels, another step towards OpenClaw parity. Anthropic’s denial of a back door into defense operations.

Machine Learning Research

A Single Tokenizer for Visual Media: Apple’s AToken, a multimodal model with a single encoder and tokenizer for images, videos, and 3D objects

Multimodal models typically use different tokenizers to embed different media types, and different encoders when training to generate media rather than classify it.

Business

DeepSeek Snubs Nvidia for Huawei: DeepSeek made its upcoming 4.0 model available for performance testing to Chinese chipmakers but not U.S, ones

DeepSeek, the Chinese developer of outstanding open-weights models, has withheld an upcoming update of its flagship model from U.S. chip makers, a move that intensifies the AI rivalry between the U.S. and China.

Alibaba's latest flagship models are open-weights MoE performers in sizes from less than 1B parameters

Machine Learning Research

Qwen3.5 Outperforms Bigger Models, Leads Vision Benchmarks: Alibaba’s latest flagship models are open-weights MoE performers in sizes from less than 1B parameters

The Qwen3.5 family of open-weights vision-language models includes impressive larger models as well as a smaller one that outperforms an OpenAI open-weights model 10 times its size.

Business

Drones Hit Persian Gulf Data Centers: Iran struck AWS facilities in Bahrain and the UAE, threatened data centers throughout the region

Iran hit at least three Amazon data centers in the Middle East, an indicator of AI’s critical role in the United States’ war against Iran and possibly the first time such facilities have been targeted during warfare.

Two cheerful robots discuss software predictions beside a large question mark at a tech expo.

The Batch Newsletter

Attacks On Data Centers, Qwen3.5 In All Sizes, DeepSeek’s Huawei Play, Apple’s Multimodal Tokenizer

The Batch AI News and Insights: I’ve been hearing from people at all levels of seniority about a feeling of job insecurity.

Letters

That Feeling Of Job Insecurity (And What You Can Do About It): In times of economic and technological uncertainty, build on things that tend to remain stable. Community and skills are likely to weather shifts in business and society.

I’ve been hearing from people at all levels of seniority about a feeling of job insecurity.

A snake crosses a busy tech startup walkway; people with scooters react in surprise.

Data Points

Nvidia’s enterprise-focused NemoClaw gives OpenClaw a security boost: Claude Dispatch lets paid users authorize remote work

GPT-5.4 Mini and Nano, optimized for speed. Nvidia’s own tiny-sized open-weights model. Midjourney updates to v8 in Alpha. Mamba releases third version of its state space language model.

Industrial facility with solar panels actively powering data servers; visible cooling towers emit steam in the sunset.

Data Points

Perplexity Computer hits desktop, mobile, Pro: Google releases Aletheia, its top natural language math agent

DeepSeek testing favors local chipmakers. Nvidia’s reasoning-added retrieval method. Why OpenAI and Oracle shut down Stargate expansion. Meta’s giant deal with AI cloud provider Nebius.

Visual schema of FAE's learning process, featuring fire and snowflake icons showing performance focus.

Machine Learning Research

Lightning-Fast Diffusion Learning: Inside Feature Auto-Encoder, a diffusion image generator that shrinks embeddings for more speed

Research shows that diffusion image generators can train somewhat faster if they learn to reconstruct embeddings from a pretrained encoder that’s built for vision tasks like classification, segmentation, and retrieval — not image generation.

Aerial view of large data centers and power plants in a desert landscape, highlighting off-grid construction.

Business

AI Data Centers Go Off the Grid: Rather than rely on public utilities, AI companies build their own power plants

Meta and OpenAI are among the tech companies that are building private power plants that will operate independently of regional grids to supply electricity for their massive buildout of AI data centers.