Welcome back!

Amazon is taking its biggest step yet into generative AI with Nova, a new suite of models for text, images, and video. With 75% cost reductions, Nova is positioned to disrupt cloud AI economics and challenge competitor prices. Let’s unpack...

Today’s Summary:

• 🚀 Amazon launches Nova AI models

• 🕹️ World Labs builds 3D walk-throughs from photos

• 🏙️ AI converts street sounds into city images

• ❓ Why does 'David Mayer' crash ChatGPT?

• 🌐 Arc is building Dia, a new AI browser

• ⚡ Amazon's Trainium 2 to power Claude

• 🎯 Cohere's new Rerank 3.5 search model

• 🛠️ 2 new tools

TOP STORY

Amazon breaks into the AI race with Nova models

The Summary: Amazon has launched Nova, a suite of AI models on its AWS cloud platform designed for text, image, and video. The lineup includes four text models and two media creation tools, offering up to 75% lower costs compared to competitors like Claude. Nova matches current state of the art AI capabilities, with plans for more advanced models in 2025, including speech-to-speech and "any-to-any" modalities.

Key details:

• Nova models run 75% cheaper than comparable AWS API options while delivering similar performance

• Supports 200+ languages and adds watermarking to generated content

• Handles up to 300,000 tokens, equivalent to a 900-page book

• Nova Reel creates 6-second videos in 3 minutes, with 2-minute videos planned for release soon

Why it matters: Amazon's entry into the AI race targets API and enterprise applications, reshaping cloud AI economics with aggressive pricing. Early adopters have raised some concerns about setup complexity and compliance. This move may pressure competitors like OpenAI, Google and Anthropic to adjust their pricing as enterprise AI adoption accelerates.

Nova API User Guide

3D AI

World Labs creates walk-through 3D worlds from photos

The Summary: World Labs has developed an AI prototype that turns regular photos into interactive 3D environments you can explore in your web browser. This early-stage preview maintains object permanence and follows physics rules, setting it apart from video-based AI models. Users can walk through the scenes, adjust camera effects, and interact with objects - though for now only within tight boundaries. The project comes from ImageNet creator Fei-Fei Li's startup, which aims to launch its first finished product in 2025.

Key details:

• Browser-based demo lets users explore AI-generated 3D scenes with keyboard and mouse controls

• Generated worlds maintain object permanence - items stay in place when out of view and then return

• System applies real-time effects like spotlight, ripples, and color waves to the 3D scenes

• Future applications include game development, movie production, and other creative industries

Why it matters: This early preview hints at future products where anyone could transform ideas into interactive 3D environments without massive development costs. It opens up opportunities for quick prototyping in game design and film.

Try the interactive demos

RESEARCH

AI turns street sounds into accurate city images

The Summary: Researchers at the University of Texas have developed an AI system that converts ambient street sounds into city images. The system learned from 10-second clips of street scenes across North America, Asia, and Europe to match ambient sounds with visual elements. It can determine time of day from traffic patterns or insect sounds, and reproduce the correct proportions of sky, buildings, and greenery in its generated images.

Key details:

• Achieved 80% accuracy in audio-to-image during human verification tests

• Trained on street recordings from three continents to capture diverse urban environments

• Detects nighttime from cricket sounds and reduced traffic noise levels

• Generates images that preserve architectural styles and building-to-sky ratios of original locations

Why it matters: This research shows how auditory and visual information are deeply connected, enabling AI to “see” places based on sound alone. Beyond its technical achievement, the system could have applications in urban planning, environmental psychology, and immersive media.

QUICK NEWS

Quick news

• Why does the name ‘David Mayer’ crash ChatGPT?

• Arc browser is developing Dia, a new AI browser

• Amazon’s new Trainium 2 chips to power Claude training

• Cohere releases Rerank 3.5 for precise AI search

TOOLS

🥇 New tools

• EasyChef - Create recipes from your ingredients using AI

• Supabase AI Assistant - Chat with your Postgres database