Good morning. It’s Wednesday, October 30th.

Did you know: On this day in 1985, Space Shuttle Challenger lifted off for its 9th flight. It was the shuttle’s last successful mission before the fatal explosion that occurred 3 months later.

In today’s email:

• GitHub’s Helps You Build AI-powered Apps

• Meta’s Llama 4 Coming Early 2025

• Anonymous Image Generator Making Waves

• 5 New AI Tools

• Latest AI Research Papers

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

GitHub's new Spark platform helps anyone build AI-powered apps with natural language

GitHub’s latest enhancements broaden Copilot’s utility by adding Anthropic’s Claude 3.5 Sonnet and Google’s Gemini 1.5 Pro to its model roster, joining OpenAI’s GPT. This update lets developers toggle between models optimized for specific coding tasks, while Copilot Enterprise users can manage model access across teams.

In addition, GitHub launched Spark, a platform for AI-driven app creation that allows users, regardless of skill level, to build micro-apps using natural language commands and integrate AI features without manual cloud management.

Other updates include Copilot’s enhanced multi-file editing in VS Code, a 30-second code review feature, and expanded Copilot Extensions support, now integrating tools like Docker and Stack Overflow. By advancing model variety and AI-driven development tools, GitHub positions itself as a hub for adaptable coding resources and setting the stage for Microsoft’s potential alignment across its broader ecosystem. Read more.

Meta Likely to Release Llama 4 Early Next Year, Pushing Towards Autonomous Machine Intelligence (AMI) ​​

Meta is preparing Llama 4 for early next year, pushing closer to autonomous machine intelligence (AMI) with real-time adaptability. Using a “Chain of Thought” approach, Meta’s VP of AI, Manohar Paluri, says Llama models can adjust dynamically—imagine rethinking a route when travel plans shift. Combining self-supervised learning (SSL) for broad knowledge and reinforcement learning from human feedback (RLHF) for precision, Llama’s capabilities continue to grow.

The recent Llama 3.2 quantized models boost performance, with faster inference speeds and lower memory demands. Meta’s AMI ambitions align with a rapid rollout cadence, with plans for Llama 4 to include upgraded memory, cross-modality, and tighter platform integration. Each update raises the bar, keeping Meta’s AI development firmly on pace. Read more.

Related Story: Meta reportedly developing custom search engine for its Meta AI chatbot

A Mysterious New Image Generation Model Has Appeared

The “red_panda” model has quietly climbed to the top of the Artificial Analysis leaderboard, outpacing heavyweights like Midjourney, Black Forest Labs’ Flux1.1 Pro, and OpenAI’s latest offerings. With an Elo score that leaves rivals in the dust, “red_panda” has gained a nearly mythical status among AI enthusiasts—particularly with its swift generation times clocking in at a breezy seven seconds, over 100 times faster than DALL-E 3.

Artificial Analysis’s ranking system, based on user-selected image evaluations, has highlighted “red_panda” not only for its output quality but also for its speed—generating images in about seven seconds, making it over 100 times faster than DALL-E 3. Despite some selection bias among the benchmark’s AI-savvy users, “red_panda” stands out. Its origin and creator remain unknown, but with labs increasingly using benchmarks to build anticipation, it’s possible an announcement about the model’s official release is imminent. Read more.

• Apple Intelligence goes live with iOS 18.1 update 

• From lightweight AI to design automation, researchers introduce advances in AI technology

• Gartner predicts AI agents will transform work, but disillusionment is growing

• ROCKET-1 mines diamonds in Minecraft by seeing and tracking objects in real time 

• xAI adds image understanding capabilities to Grok

• We finally have an 'official' definition for open source AI

• VistaDream: Sampling multiview consistent images for single-view scene reconstruction

• DepthSplat: Connecting Gaussian Splatting and Depth

• Inside Elon Musk's Colossus: xAI's World-Leading AI Supercomputer

• US finalizes rules to curb AI investments in China, impose other restrictions

• Google’s AI search summaries are rolling out to over 100 more countries

• ​​ChatGPT will let you search your chat history on the web

• OpenAI builds first chip with Broadcom and TSMC, scales back foundry ambition

• Meta Introduces MarDini, its New Family of Video Diffusion Models

• Stability AI releases Stable Diffusion 3.5 Medium with lower VRAM requirements

• Google CEO teases the company’s next Gemini model

• LinkedIn launches its first AI agent to take on the role of job recruiters

• Generative AI could generate millions of tons of e-waste by decade's end, study finds

• Study finds LLMs can identify their own mistakes

• Amazon Q introduces AI-powered inline chat directly in the code editor for developers

• Elon Musk's xAI is reportedly trying to raise billions more

• ​​Nvidia hands out blueprints for the creation of scalable 'AI factories'

• SAG-AFTRA has announced an agreement with Ethovox, which is building a “foundational AI model for voice.”

📄 AutoKaggle: A Multi-Agent Framework for Autonomous Data Science Competitions

📄 LongReward: Improving Long-context Large Language Models with AI Feedback ​​

📄 Continuous Speech Synthesis using per-token Latent Diffusion

📄 CLEAR: Character Unlearning in Textual and Visual Modalities

📄 AgentStore: Scalable Integration of Heterogeneous Agents As Specialized Generalist Computer Assistant

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 or DM us on X!