Good morning. It’s Monday, July 29th.

Did you know: Astronomers announce the discovery of dwarf planet Eris, leading the International Astronomic Union to clarify the definition of a planet.

In today’s email:

• OpenAI and Anthropic face huge financial losses

• Apple's AI features are delayed to October due to EU regulations

• China's AI is advancing rapidly using U.S. open-source tech

• A study reveals AI struggles with abstract visuals

• Anthropic is accused of excessive data scraping

• 5 New AI Tools

• Latest AI Research Papers

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

OpenAI and Anthropic Loosing Billions?

An analysis by The Information reveals significant financial losses for AI companies OpenAI and Anthropic. OpenAI faces potential losses up to $5 billion this year, with expenses related to training AI models and running inference systems estimated to reach $7 billion. The costs include around $4 billion for Microsoft server rentals and $3 billion for training, with personnel expenses projected at $1.5 billion. Anthropic also reports substantial losses, forecasting over $2.7 billion in spending against estimated revenues of $800 million.

The industry is grappling with a fundamental question: can these AI models, despite their impressive feats of mimicry, generate enough tangible value to justify their astronomical costs? While there's potential for growth in new product areas, the current state of affairs suggests that AI may need a significant upgrade in its cognitive abilities—perhaps a breakthrough in general reasoning—to truly justify the exorbitant price tag. Until then, the AI industry finds itself in a precarious position, balancing on a tightrope between ambition and economic reality. Read more.

Apple's AI features to be delayed until October

Apple’s planned AI features, branded as Apple Intelligence, will be delayed and are not expected to launch with the initial iOS 18 and iPadOS 18 releases in September. According to Bloomberg News, these new AI capabilities will roll out a few weeks later, likely in October. Developers will receive early access through iOS 18.1 and iPadOS 18.1 betas starting this week.

Apple Intelligence will enhance device functionalities by generating text, images, and other content on demand. It will be available on the iPhone 15 Pro, iPhone 15 Pro Max, and iPads and Macs with M1 chips or later. This delay is due to Apple’s need to comply with European Union tech regulations that require compatibility with rival products. Read more.

China's AI models soar with help from US open-source technology

China's AI surge, fueled by U.S. open-source tech, is rapidly closing the gap with Western models. Companies like Kuaishou and 01.AI are leading the charge, producing AI systems that rival U.S. benchmarks. However, the open-source nature of underlying technology complicates regulation, potentially accelerating Chinese advancements.

Unlike their Western counterparts, Chinese firms are more aggressive in deploying consumer-facing AI. Cultural considerations also shape AI development, with models aligning to socialist values. Geopolitical tensions are evident as the U.S. imposes chip export restrictions to slow China's progress, and OpenAI limits service to authoritarian states. Read more.

Study reveals major weaknesses in AI's ability to understand diagrams and abstract visuals

A recent study exposed a critical blind spot in AI's visual intelligence. While these models excel at interpreting photographic realism, they falter dramatically when presented with abstract visualizations like diagrams and charts. Advanced models such as GPT-4.0 and Claude 3.5 Sonnet achieved a mere 64.7% and 59.9% accuracy respectively on a diverse dataset of 11,193 abstract images, significantly lagging human performance at 82.1%. The research team employed a "multimodal self-instruct" approach to develop a diverse dataset of 11,193 abstract images, each accompanied by specific questions across eight scenarios, such as dashboards, road maps, diagrams, tables, and more. Despite their multimodal capabilities, the models struggle particularly with tasks requiring spatial reasoning and abstract concept understanding. Attempts to improve open-source model performance using synthetic data showed some success, yet reliance on costly closed models for reference data poses challenges. Read more.

AI start-up Anthropic accused of ‘egregious’ data scraping

AI start-up Anthropic faces accusations of excessive data scraping, which may contravene publishers' terms of service. Web publishers, including Freelancer.com and iFixit.com, have reported substantial traffic spikes from Anthropic's web crawlers, with some sites receiving millions of hits in short periods.

Despite efforts to block or limit these crawlers, such as using standard web protocols, the scraping continued, prompting complaints about disruption and increased operational costs. Anthropic asserts it respects protocols like 'robots.txt' and aims to avoid being intrusive. The practice of scraping, while legally permissible, has intensified with the growing demand for data to train large language models. This has led to increased bandwidth costs and operational strain on affected sites. Read more.

Down the Rabbit Hole:

New AI models flood the market even as AI takes fire from regulators, actors and researchers

Artificial Intelligence Will Let Humanity Talk to Alien Civilizations

Computer love: AI chatbots are changing how we understand romantic and sexual well-being

Llama 3.1 vs GPT-4o vs Claude 3.5: A Comprehensive Comparison of Leading AI Models

Brev.ai uses AI to generate music from text, offering diverse genres and online access, making music creation user-friendly.

ZETIC.MLange enables efficient on-device AI model execution using NPUs, supporting multiple platforms like Android and iOS for seamless integration.

PixVerse V2 allows users to create 8-second AI-generated videos with enhanced detail and motion for consistent storytelling across multiple clips.

Immersity AI uses depth maps to convert 2D media into immersive 3D content, enhancing user experiences on XR devices.

MarsCode offers an AI-integrated cloud IDE with code completion, debugging, schema management, and one-click deployment for enhanced development efficiency.

📄 Diffree: Text-Guided Shape Free Object Inpainting with Diffusion Model

📄 LAMBDA: A Large Model Based Data Agent ​​

📄 Very Large-Scale Multi-Agent Simulation in AgentScope  

📄 AppWorld: A Controllable World of Apps and People for Benchmarking Interactive Coding Agents

📄 AMEX: Android Multi-annotation Expo Dataset for Mobile GUI Agents

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 with “Sponsorship” in the subject line.