Good morning. It’s Friday, July 5th.

Did you know: On this day in 1996, Hotmail was introduced. It would later be acquired by Microsoft for around $400 million and rebranded as "MSN Hotmail"

In today’s email:

• Figure Robot

• Synthetic Data Army

• Suno App

• 5 New AI Tools

• Latest AI Research Papers

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

Figure 01: Coffee-making humanoid robot now shows car assembly skill at BMW

Figure AI, a humanoid robotics company, deployed its Figure 01 model at BMW's Spartanburg, South Carolina facility in January 2024. Recently, a video was released showcasing the robot's role in vehicle assembly, with its neural networks translating visual data into actions.

The fully electric Figure 01 stands at 1.6 meters tall and 60-kilogram weight with a 20-kilogram payload capacity for a five-hour shift. Over the next one to two years, the robots will undergo further training for specific tasks. Utilizing advanced processing from OpenAI, the robot can process visual information at 10 Hz and execute actions at 200 Hz, enabling precise manipulation and navigation.

Companies like Honda, Hyundai, and Mercedes are also exploring similar technologies to automate repetitive and potentially hazardous tasks. Read more.

Tencent researchers unleash an army of AI-generated personas for data generation

Tencent AI Lab in Seattle has launched a new method for synthetic data generation using AI-driven personas, termed "personas," to improve the creation of diverse datasets essential for training AI systems. These personas, totaling over a billion in the Persona Hub, are generated through two main techniques: "Text-to-Persona," extracting personalities from web texts, and "Persona-to-Persona," generating new personas based on existing relationships. This dual approach allows for the generation of varied datasets tailored to specific AI training requirements, similar to how different prompts influence outputs in language models.

AI-generated personas play critical roles in generating datasets for training AI models in various domains like mathematical problem-solving, logical reasoning, and language processing. For example, Tencent researchers utilized the Persona Hub to create 1.07 million math problems, achieving high accuracy with smaller model sizes compared to traditional methods.

This research suggests a potential shift in AI training data generation. In the future, large language models may leverage this approach to generate diverse synthetic data independently, potentially reducing reliance on human-sourced data. Read more.

The future of music creation? Suno's AI app hits the App Store

Suno, known for its AI-powered text-to-music technology, has introduced its iOS app, offering a virtual music studio accessible to millions. Users can create original songs by entering text prompts, with Suno's AI generating full compositions including vocals and instruments in minutes.

This launch is a big step for AI music creation as it aims to make production accessible to everyone. However, it also raises questions about creativity and copyright. Despite legal challenges, including copyright lawsuits from major labels, Suno remains optimistic about its technology's impact. Read more.

Etcetera: Stories you may have missed

Salesforce proves less is more: xLAM-1B 'Tiny Giant' beats bigger AI Models

OpenAI’s ChatGPT Mac app was storing conversations in plain text

VALL-E 2: Microsoft's new AI voice tech is so good they're afraid to release it

Perplexity AI's new Pro Search can tackle complex queries, but faces scrutiny over data practices

Meta open-sources new ‘multi-token prediction’ language models

French AI lab Kyutai unveils conversational AI assistant Moshi, plans open-source release

Google's ImageInWords could boost everything from image search to text-to-image AI

Tesla shows its humanoid robot Optimus at China AI conference, but behind glass

OpenAI's internal AI details stolen in 2023 breach, NYT reports

Whiteboard of Thought: New method allows GPT-4o to reason with images

WhatsApp is developing a personalized AI avatar generator

Japan introduces enormous humanoid robot to maintain train lines

Top Biden administration officials withdraw from Coalition for Health AI

Morehouse to be the first university to use virtual AI teaching assistants

DARPA wants to use AI to find new rare minerals

Hubflo enhances client interactions with an AI-built client portal featuring file sharing, contracts, messaging, tasks, billing, and integration with 500+ tools.

Archivist offers a fast semantic search for codebases, indexing and querying code in milliseconds, helping developers quickly navigate and understand large codebases.

Mammouth integrates Midjourney, Dall-E, and Stable Diffusion with top LLMs, offering the latest versions under one subscription for seamless AI access.

Typelets offers cloud-based file editing enhanced by Generative AI, facilitating productivity and collaboration with robust, feature-rich tools for seamless user experiences.

Recroo AI streamlines hiring with an end-to-end AI interview app, offering interactive video interviews, AI-powered screening, and unbiased feedback reports.

📄 Summary of a Haystack: A Challenge to Long-Context LLMs and RAG Systems

📄 InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output ​​

📄 OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation

📄 Agentless: Demystifying LLM-based Software Engineering Agents

📄 TokenPacker: Efficient Visual Projector for Multimodal LLM

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, apply here.