Good morning. It’s Monday, August 26th.

Did you know: On this day in 2011, Steve Jobs stepped down as Apple CEO due to health problems.

In today’s email:

• Meta’s Foundation for Human Models

• Google’s Prompt Gallery

• Grok-2 Rewrite

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with}

Ready to Scale Your Business?

CodeMode is your development partner to:

Boost Your Web Presence: We create powerful websites, apps, and software that engage your audience and fuel growth.

Top the Search Rankings: Our advanced SEO strategies will drive more qualified traffic to your site.

Streamline Operations: Automate with AI-driven solutions, from smart document processing to chatbots.

Focus on what you do best — leave the tech to us.

Meta Presents Sapiens: Foundation for Human Vision Models

Meta's "Sapiens" introduces a new approach to human-centric computer vision through vision transformer models pretrained on a vast dataset of human images. Addressing a limitation in current models that mainly focus on general image pretraining, Sapiens utilizes the Humans-300M dataset—comprising 300 million human images—to improve performance in specific tasks like 3D digitization, keypoint estimation, and body-part segmentation.

The models feature large-scale pretraining and high-resolution vision transformers, leading to evident performance gains. Evaluations show enhanced results in pose estimation, part segmentation, depth estimation, and normal estimation. Sapiens’ models are effective in generalizing across various scenarios and offer a solid foundation for future research and practical applications in human-centric vision. Read more.

Related Story: Meta cancels plans for high-end mixed-reality headset, the Information reports: Meta has canceled its plans for a high-end mixed-reality headset, which was intended to compete with Apple’s Vision Pro. The decision followed a recent product review meeting, leading Meta's Reality Labs division to cease development on the project, code-named La Jolla. This headset was slated for a 2027 release and was expected to feature micro OLED screens, similar to those used in the Vision Pro. Read more.

Google debuts free 'Prompt Gallery' in AI Studio, supercharging developer tools

Google has introduced a new feature called Prompt Gallery in its AI Studio platform, significantly expanding the tools available for developers using the Gemini API. Announced by Logan Kilpatrick, the Prompt Gallery provides a range of pre-built prompts that highlight the versatility of Google’s Gemini models.

These include practical tools such as a JSON schema-based recipe generator, a math tutor for quadratic equations, and a time complexity analyzer, as well as creative applications like a scavenger hunt creator and a trip recommendation system. This update enhances AI Studio by supporting long context, multi-modal inputs, and structured outputs. Read more.

Grok-2 gets a speed bump after developers rewrite code in three days

Developers Lianmin Zheng and Saeed Maleki have dramatically improved the performance of Elon Musk’s xAI Grok-2 large language model (LLM) and its smaller counterpart, Grok-2 mini, through a complete rewrite of the inference code. Over the past three days, they utilized SGLang, an efficient open-source system, which boosted throughput by up to 6.4 times compared to previous methods.

As a result, both models now deliver faster analysis and response times, with Grok-2 mini’s speed doubling and the main Grok-2 model also benefiting from enhanced performance and slightly improved accuracy. The Grok-2 model now holds the #2 position on the LMSYS Chatbot Arena leaderboard, while Grok-2 mini ranks #5. Read more.

Etcetera: Stories you may have missed

Stephen Wolfram thinks we need philosophers working on big questions around AI

Google gives people back the voice they never had

Anthropic’s CEO has some thoughts about California’s AI safety bill, too.

AI Models Complex Molecular States with Precision

Nvidia to present AI and data center performance innovations at Hot Chips conference

GitHub named a Leader in the Gartner first-ever Magic Quadrant for AI Code Assistants

Report: Chinese organizations use public cloud to access restricted AI chips

📄 LLM Pruning and Distillation in Practice: The Minitron Approach

📄 Controllable Text Generation for Large Language Models: A Survey  ​​

📄 LayerPano3D: Layered 3D Panorama for Hyper-Immersive Scene Generation

📄 MEDCO: Medical Education Copilots Based on A Multi-Agent Framework

📄 TWLV-I: Analysis and Insights from Holistic Evaluation on Video Foundation Models  

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email!