Another batch of AI robots is about to clock in at the factory...

🌏 Figure AI partners with BMW

🌟 Meta 3D Gen launches

🤝 Apple: New OpenAI board observer

💼 Microsoft open-sources GraphRAG

📈 "Space Computing Constellation" construction begins

🧠 Moonshot AI rolls out context caching

💰 Runway valued at $4 billion in latest round

📊 Benchmark 1 raises $425 million

👁️ KP's $2 billion new fund targets AI

🌱 Sentient closes $85 million seed round

🌌 A Midsummer Night's Dream of LLM exploration

---

Figure AI partners with BMW

Another batch of AI robots heading to the factory floor…

Figure AI has reached a groundbreaking partnership with BMW Group to bring humanoid robot technology into automotive manufacturing, signaling a new era of industrial automation. Figure's robots are designed to handle diverse tasks — from complex precision operations to repetitive labor — with the goal of boosting efficiency and safety on production lines. The first phase of the collaboration will identify potential applications for robots in production, followed by gradual deployment at BMW's South Carolina plant.

The two sides will also conduct in-depth research on AI and robotics integration technologies, jointly shaping the future of intelligent manufacturing. Since its founding in 2022, Figure AI has focused on developing general-purpose humanoid robots capable of working independently in real-world environments, addressing labor shortages while improving production efficiency. The company has secured backing from industry leaders including OpenAI and Microsoft, and completed a $750 million funding round in 2023 at a $2.2 billion valuation. Despite competition from companies like Boston Dynamics and Tesla, Figure AI shows tremendous potential in advancing humanoid robot technology.

Meta 3D Gen launches

Meta 3D Gen[1] (3DGen for short) is cutting-edge technology from Meta that provides a fast pipeline for text-to-3D asset generation. 3DGen can generate high-quality 3D shapes and textures with high fidelity in under a minute, supporting physically-based rendering (PBR) — critical for relighting 3D assets in real-world applications. Additionally, 3DGen can perform retexturing of previously generated (or artist-created) 3D shapes using additional text input provided by the user.

3DGen integrates two key technical components: Meta 3D AssetGen and Meta 3D TextureGen, for text-to-3D and text-to-texture generation respectively. By combining the strengths of both technologies, 3DGen can represent 3D objects simultaneously in three ways: view space, volumetric space, and UV (or texture) space. This technological integration achieves a 68% win rate when compared against single-stage models.

Apple: New OpenAI board observer

Apple has secured an observer seat on OpenAI's board — a strategic move signaling deeper cooperation between the two companies. Phil Schiller, Apple's former marketing chief and current App Store lead, will take on this role. While he won't have voting rights, he can attend meetings and gain insight into the company's decision-making core. This arrangement is expected to take effect within the year.

A rundown of Apple and OpenAI's strategic partnership: In June, Apple announced it would integrate ChatGPT into iPhone, iPad, and Mac as part of its AI features. The Apple-OpenAI deal isn't a financial transaction — Apple promotes OpenAI's technology in exchange for access. Apple believes that bringing OpenAI's technology to hundreds of millions of devices is worth more than a cash payment.

Apple is also in discussions with Google, Anthropic, Baidu, and Alibaba about bringing additional chatbot capabilities to its platforms.

Microsoft open-sources GraphRAG

Microsoft has officially open-sourced GraphRAG[2] on GitHub — a knowledge graph-based retrieval-augmented generation (RAG) method for question-answering over private datasets. GraphRAG provides more structured information retrieval and comprehensive response generation compared to traditional RAG approaches. It uses large language models (LLMs) to automatically extract knowledge graphs from text documents and create hierarchical summaries of data, providing an overview of the dataset. The GraphRAG code repository includes a solution accelerator that simplifies deployment on Azure.

Compared to standard RAG, GraphRAG delivers significant performance improvements in question-answering, particularly for complex questions requiring synthesis of multiple information points. It builds community hierarchies and generates summaries, then leverages these structures to provide contextual material to LLMs when executing RAG-based tasks to answer questions.

"Space Computing Constellation" construction begins

Zhejiang Lab and Chengdu Guoxing Aerospace Technology Co. (国星宇航) have signed a strategic cooperation agreement to jointly advance construction of the "Space Computing Constellation." The constellation aims to use carbon-free energy in space to build a space-based computing network, addressing the high costs of ground-based computing power. Compared to traditional ground computing, the Space Computing Constellation has inherent advantages, capable of reducing data processing latency and network data transmission volume.

Guoxing Aerospace plans to launch an AI computing satellite in August that will serve as a space computing node in the space-based computing network, with a focus on in-orbit validation of multimodal data processing capabilities. Previously, Guoxing Aerospace successfully launched and validated the world's first AI on-orbit super-resolution commercial satellite, equipped with its self-developed "Satellite Intelligent Brain," completing in-orbit validation of 4x AI super-resolution algorithms. This cooperation will lay important groundwork for enhancing space computing service capabilities and constellation networking.

Moonshot AI rolls out context caching

Beijing Moonshot AI Technology Co. has introduced an innovative context caching technology on its Kimi platform, which can significantly reduce developers' costs for using long-context large models — by up to 90%. By pre-storing frequently requested data, the technology allows the system to quickly serve information from cache, avoiding repeated computation and data retrieval processes.

In one example, the company demonstrated how context caching technology reduced the cost of a Q&A session that originally required processing large volumes of text from ¥141.95 to ¥11.88 using conventional methods. This breakthrough provides Moonshot AI with competitive advantage in the current market where large model providers are fiercely competing, and could reshape pricing models for AI large model services.

Runway valued at $4 billion in latest round

Runway has completed a new $450 million funding round led by General Atlantic[3], with its valuation soaring to $4 billion. This valuation reflects its breakthroughs in AI video generation technology, particularly the recently released Gen-3 Alpha model, which improved controllability and speed. Despite a gap with industry giants like OpenAI, Runway's rapid growth and professional market positioning have won market recognition. Additionally, Runway actively engages in community initiatives to alleviate industry concerns about potential employment impacts from AI technology.

Benchmark 1 raises $425 million

Benchmark Capital[4], focused on generative AI, plans to raise a new $425 million fund called "Benchmark 1." The firm has stayed consistent, never expanding its scale even when VC funding was abundant in 2021, maintaining six partners. Benchmark's strategy is now seen as a wise guard against market bubbles. Its recent investments include AI startups HeyGen and Sierra, led by former Salesforce CEO Bret Taylor.

KP's $2 billion new fund targets AI

Venture capital firm Kleiner Perkins[5] announced two new funds: KP21 and KP Select III, sized at $825 million and $1.2 billion respectively. KP21 is designed to support startups, while KP Select III focuses on high-impact investments. In its statement, the firm emphasized that these funds will continue to focus on enterprise software, consumer, healthcare, fintech, and hard tech. And in every category, KP will pay special attention to how these companies will use AI.

KP partner Ilya Fushman noted: "AI is an accelerant. AI may be one of the biggest accelerants and economic opportunities in tech history."

Sentient closes $85 million seed round

Sentient[6], an open-source AI platform, announced it has completed an $85 million seed funding round. The capital will go toward platform development and engineering team building. Sentient aims to help AI developers monetize open-source models and data, pushing back against the current dominance of closed-source AI. Backers behind the project include Polygon founder Sandeep Nailwal, Princeton University professor Pramod Viswanath, and Himanshu Tyagi of the Indian Institute of Science.

Sentient will use blockchain protocols and incentive mechanisms to ensure developers' economic interests, driving the development of open AI. Founders Fund partner Joey Krug emphasized the importance of Sentient addressing the incentive problem for open-source AI adoption. Sentient's testnet is planned for launch in Q3 2024. In addition to Founders Fund, investment firms including Robot Ventures, Symbolic Capital, and Dao5 participated in this round.

A Midsummer Night's Dream of LLM exploration

The internet's classic large model learning "textbook" — Large Model Open Course — returns for Season 2.

The globally renowned open-source communities OpenBMB and Hugging Face have joined forces in a dream collaboration, with Crossing also participating as a community partner, to craft a large model learning journey for AI explorers, setting sail on July 10.

On this midsummer night, the secret to a happy dream lies not only in the gentle evening breeze, but surely also in a satisfied curiosity. Whether you want to learn the basics of building LLMs or absorb the latest research from AI talent scouts, you'll find your answers here. Please click "read more" for details.

You cannot choose the value you have in the eyes of others, so you must recognize that doing the right thing is enough, whether or not anyone notices.

— John McCarthy, AI pioneer and Turing Award winner

It was he who originally proposed the Dartmouth Conference in 1956 and coined the term "artificial intelligence."

Editor: Yuki Design: Ivan

For business cooperation, add on WeChat: Rwkfbcianvd

References

[1] Meta 3D Gen: https://ai.meta.com/research/publications/meta-3d-gen/

[2] GraphRAG: https://github.com/microsoft/graphrag

[3] General Atlantic: https://www.generalatlantic.com/

[4] Benchmark Capital: https://www.benchmark.com/

[5] Kleiner Perkins: https://www.kleinerperkins.com/

[6] Sentient: https://sentient.foundation/