OpenAI + humanoid robots 🔥

In today’s email:

👀 The 9-Month-Old AI Startup Challenging Silicon Valley’s Giants
☠️ Malicious AI models on Hugging Face backdoor users’ machines
🤩 How Singapore is preparing its citizens for the age AI.
🧰 9 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI + humanoid robots

Figure, an AI robotics firm based in Sunnyvale, California, recently secured $675 million in Series B funding, achieving a valuation of $2.6 billion. The funding round saw contributions from major investors including Microsoft, NVIDIA, Jeff Bezos through Bezos Expeditions, and several venture capital firms. This influx of capital is set to boost Figure's efforts in developing general-purpose humanoid robots, aiming for accelerated commercial deployment.

Additionally, Figure has partnered with OpenAI to create advanced AI models for these humanoid robots, combining OpenAI's AI research prowess with Figure's robotics expertise. This collaboration is expected to enhance the robots' language processing and reasoning abilities, contributing to Figure's goal of deploying humanoid robots in commercial settings. To support its AI and robotics development, Figure will utilize Microsoft Azure for its AI infrastructure, benefiting from Azure's robust training and storage capabilities.

The company, which boasts a skilled team from renowned organizations like Boston Dynamics and Tesla, is focused on bringing humanoid robots into various industries to address labor shortages and improve efficiency. Figure's recent agreement with BMW Manufacturing to integrate humanoid robots into automotive production highlights the practical applications of their technology. With the new funding and strategic partnerships, Figure is poised to advance the role of AI and robotics in the workforce, promising a significant impact on various sectors.

Malicious AI models on Hugging Face backdoor users’ machines

JFrog's security team discovered about 100 malicious AI and machine learning models on the Hugging Face platform, capable of executing code on users' machines to create backdoors. Despite Hugging Face's extensive security protocols, these models could potentially lead to data breaches and espionage. The malicious models were identified through an advanced scanning system targeting PyTorch and Tensorflow Keras models, emphasizing the threat they pose to the platform's users.

In one notable instance, a PyTorch model uploaded by a user contained a payload that established a reverse shell, allowing external command execution. This malicious activity was concealed within the model's serialization process, demonstrating the sophistication of these threats. JFrog's analysis suggests some of these uploads might be from AI researchers testing security boundaries, not just malicious hackers.

JFrog's findings serve as a warning about the potential security risks associated with AI and ML models. The incident underscores the need for heightened vigilance and stronger protective measures within the AI community to prevent the exploitation of platforms like Hugging Face by malicious actors.

Meta plans launch of new AI language model Llama 3 in July

Meta Platforms is gearing up to launch its latest artificial intelligence model, Llama 3, in July, aimed at enhancing user interaction with more nuanced responses to complex inquiries, as reported by The Information. This new version is part of Meta's initiative to refine its AI's understanding and response mechanism, especially when dealing with sensitive or contentious issues, allowing it to provide contextual insights rather than outright refusing to engage.

The development of Llama 3 is seen as a significant step forward from its predecessor, Llama 2, which currently powers chatbots across Meta's social media platforms but has limitations in addressing certain types of questions. By enabling Llama 3 to interpret and respond to queries more effectively, Meta aims to improve user experience and maintain a competitive edge in the rapidly evolving AI landscape, particularly in light of Google's recent pause on its AI's image-generation feature due to accuracy concerns.

As part of the rollout of Llama 3, Meta is also planning to enhance its internal oversight on the model's training, focusing on tone and safety to ensure that the AI's responses are not only more intelligent but also aligned with ethical guidelines. This move underscores Meta's commitment to responsible AI development, balancing technological advancement with user safety and trust.

Google brings Stack Overflow’s knowledge base to Gemini for Google Cloud

Stack Overflow is introducing a new program called OverflowAPI, allowing AI companies to access its knowledge base, with Google as the launch partner. This collaboration aims to integrate Stack Overflow's validated answers into Google Cloud and introduce AI-enhanced features on Stack Overflow's platform. The partnership, which will be showcased at Google's Cloud Next conference, underscores the growing trend of leveraging expert-driven content for AI model training, albeit without revealing the financial specifics of the deal.

The partnership's vision extends beyond AI, with plans to integrate Stack Overflow directly into the Google Cloud console, enabling developers to access and query Stack Overflow's database alongside Google-specific information. This integration aims to streamline the developer experience by providing a unified source of reliable information, enhancing productivity without the need to navigate multiple platforms.

On Stack Overflow's end, the collaboration with Google's Gemini through the Vertex AI platform will potentially introduce AI support in various site functionalities, such as question-asking and moderation. This move is designed to maintain the high-quality, expert-driven content Stack Overflow is known for while adapting to the evolving landscape of AI in developer workflows, ensuring a blend of human expertise and AI efficiency.