Welcome back!

Kuaishou, a major competitor to TikTok, has just launched Kling, a cutting-edge AI video generator that rivals OpenAI’s forthcoming Sora. With the ability to create 1080p videos up to two minutes long, Kling and similar models like Google Veo are set to revolutionize how video content is produced. Let’s unpack this...

Today’s Summary:

• 🎬 Kuaishou debuts Kling Video AI generator

• 🤖 Qwen2 challenges Meta Llama-3

• 🌊 AI aids coral reef conservation

• 🧠 16 million concepts found in GPT-4

• 🛡️ Antitrust inquiries target AI giants

• ⚖️ Meta faces complaints over training models with user data

• 🛠️ 2 new tools

TOP STORY

Kling: China answer to OpenAI Sora video generator

The Summary: Chinese tech giant Kuaishou, Tiktok main competitor, has introduced Kling, a new AI model that generates high-quality videos from text prompts. Kling can produce 1080p videos up to two minutes long, with complex motions and realistic physics. The AI is set to compete directly with the forthcoming OpenAI Sora and Google Veo.

Key details:

• Kling can generate videos at 1080p resolution and 30fps

• Videos can be up to two minutes long

• Uses a 3D space-time attention system for better motion modeling

• Employs a diffusion transformer to merge concepts and create fictional scenes

• Currently on a waitlist for Chinese users only (requires a Chinese phone number)

• Kuaishou is mainly known for its social media apps and is TikTok’s main competitor

Why it matters: The introduction of Kling shows that the most advanced AI video technology is not limited to Western tech companies. As this technology continues to advance, it could transform the film, marketing, and media industries globally.

MODELS

Qwen2 new open-source model challenges Meta Llama 3

The Summary: Qwen2, the latest open-source AI model from Alibaba, significantly outperforms its predecessor Qwen 1.5 and strongly rivals Meta’s Llama-3 70B in several benchmarks. With five models ranging from 0.5B to 72B parameters, Qwen2 scores very high in natural language understanding, coding, and mathematics. It supports 29 languages and handles contexts up to 128K tokens, far surpassing Llama 3’s 8K tokens. Qwen2’s instruction-tuned models, especially the 72B variant, demonstrate high performance across benchmarks.

Key details:

• Five models: Qwen2-0.5B to Qwen2-72B

• Supports 29 languages

• Handles context lengths up to 128K tokens

• Qwen2 72B outperforms Llama-3 70B in multiple benchmarks

• Excels in coding (HumanEval: 86.0%) and math (MATH: 59.7%)

• Apache 2.0 license for most models

Why it matters: Qwen2 represents a significant development in open-source AI, making high-performance, multilingual AI freely accessible. The adoption of the Apache 2.0 free license for most models will allow the community to build new advancements and applications.

ENVIRONMENT

AI listens to save coral reefs

The Summary: Google DeepMind, in collaboration with marine biologists, has created SurfPerch, an AI tool that analyzes coral reef sounds to assess reef health and biodiversity. SurfPerch processes thousands of hours of underwater audio to monitor reef conditions. The AI learns from public contributions on the captivating Calling in our Corals website, where people can identify fish sounds from recordings. This data is then used as training data for the AI, which accelerates conservation efforts, by providing crucial insights into reef health.

Key details:

• Coral reefs cover 0.1% of ocean surface but host 25% of marine species

• Project invites the public to listen to reef audio

• SurfPerch AI is trained on contributor data to detect new reef sounds

• Bird recordings unexpectedly improved SurfPerch performance

• Initial trials showed differences in Philippine reefs and Indonesian restorations

• The public can contribute by listening to new audio on the website

Why it matters: Coral reefs are among the most vulnerable ecosystems on Earth, facing threats like overfishing, diseases, and climate change. SurfPerch provides a fast, efficient way to monitor reef health, even in deep or murky waters. AI-driven monitoring could accelerate conservation efforts, helping protect these vital ecosystems that support a quarter of all marine life.

QUICK NEWS

Quick news

• 16 million concepts have been identified in GPT-4

• US government clears way for antitrust inquiries into Nvidia, Microsoft and OpenAI

• Meta faces complaints in 11 countries over data use to train AI

TOOLS

🥇 New tools

• DuckDuckGo AI Chat - Privacy focused AI Chat

• Trustguide - Check the reputation of any company