The AI Rundown by Lightscape Partners - 07/08/24

Good morning, Happy Fourth of July to those who celebrated, and welcome back to another week of AI news.

Last week in AI:

• Cloudflare released a new feature to help protect from unsolicited web scraping for AI model training.

• Despite Google’s climate footprint goals, the tech company has increased emissions at a fast rate as the demand for model training and data centers increases.

• While the US has tried to limit China from acquiring Nvidia GPUs, an underground market for the chips has popped up.

Keep reading to see more from last week.

If you haven’t yet, please support the newsletter by subscribing!

Top AI Stories of the Week

Cloudflare releases a new feature blocking web scrapers from taking data for model training. Link.

• Cloudflare has introduced a new feature to block AI bots, scrapers, and crawlers with a single click, available to all customers, including free-tier users.

• This tool aims to protect original content from being used by AI companies without consent.

• The feature responds to the increasing demand for web data to train AI models and targets both transparent and deceptive AI bots.

• Cloudflare's machine learning models detect and block bots, even those attempting to disguise their activities.

Google’s emissions climb 50% as data center energy demands increase. Link.

• Google's goal of reducing its climate footprint is at risk due to the rising energy demand from its AI-driven data centers.

• Over the past five years, the tech giant's greenhouse gas emissions have increased by 48%, primarily due to electricity consumption by data centers and supply chain emissions.

• In 2023 alone, emissions rose by 13% to 14.3 million metric tons.

• As AI continues to grow, data centers are projected to consume up to 4.5% of global energy generation by 2030, complicating efforts to meet climate goals.

Underground Market Circumvents U.S. Export Controls on Nvidia AI Chips. Link.

• An underground network bypasses U.S. export controls on Nvidia AI chips, involving buyers, sellers, and couriers smuggling chips into China.

• Many Chinese distributors advertise Nvidia chips online and maintain a steady supply through preorders and discreet shipments.

• Despite U.S. restrictions, Chinese AI companies, research institutions, and universities continue to acquire these chips due to their critical role in AI development.

• The ongoing demand and limited domestic alternatives keep the market for Nvidia's high-end chips robust.

Robotics

A new training technique for AI robotics uses sounds. Link.

• AI-powered robots can now be trained using sound, helping them perform tasks in low-visibility environments.

• Stanford researchers developed a system that uses audio data to teach robots household tasks, significantly improving their performance. For example, using sound, a robot's accuracy in detecting dice in a cup increased from 27% to 94%.

• This research allows for scalable, real-world audio collection, enhancing robot training.

• Despite its benefits, audio training has challenges, such as dealing with motor noise and soft objects, which researchers aim to address by expanding data collection methods.

AI Models

French AI Lab Kyutai Releases 'Moshi,' a Competitor to OpenAI's GPT-4o. Link.

• French AI lab Kyutai has launched Moshi, an open-source, real-time multimodal AI model built on the Helium 7B model.

• Moshi features text and audio training, can express 70 emotions, and handle two audio streams simultaneously. It supports multiple backends and runs on consumer-grade hardware, making it highly accessible.

• Developed in just six months by eight researchers, Moshi aims to revolutionize human-machine communication.

• The lab plans to release the full model, challenging major AI companies by promoting open research in AI.

ElevenLabs Brings Iconic Voices to AI Reader App. Link.

• ElevenLabs has partnered with the estates of Judy Garland, James Dean, Burt Reynolds, and Sir Laurence Olivier to feature their voices in the ElevenLabs Reader App.

• This app converts digital text into emotionally rich, context-aware voiceovers using advanced AI technology.

• This marks a significant step in ElevenLabs' mission to provide high-quality AI-generated audio content.

RunwayML Seeks $450M Investment and Launches Gen-3 Alpha. Link.

• RunwayML aims to secure a $450 million investment led by General Atlantic, valuing the company at $4 billion.

• They have introduced Gen-3 Alpha, an AI model capable of generating high-quality 10-second videos, which is costly at $1 per video.

• Despite the high costs, the model excels in real-life simulations and establishing shots. RunwayML has raised $237 million so far, with backing from notable investors like Google, Nvidia, and Salesforce Ventures.

• The AI video market is becoming increasingly competitive with multiple players and substantial venture capital interest.

AI Data Infrastructure Market Map

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].