Sponsored by

ChatGPT gets a body

Two weeks ago, Figure and OpenAI announced their collaboration to advance robot learning, developing cutting-edge AI models for humanoid robots. They've made significant progress, eliminating the need for teleoperation and achieving a notable speed increase in their robots, now nearing human pace. Their approach utilizes end-to-end neural networks, with the robots' actions driven by complex algorithms that interpret data from onboard cameras and execute 24 degrees of freedom actions at a remarkable speed.

Figure's innovation extends beyond AI, as they've taken a comprehensive approach by also designing and integrating various components like motors, firmware, and battery systems. Their collaboration with OpenAI has enabled their robot, Figure 01, to not only describe its visual experiences and plan actions but also engage in full conversations, reflecting on its memory and articulating its reasoning. This is powered by a sophisticated vision-language model that processes both images and speech, enhancing the robot's interaction with its environment and decision-making processes.

The technical advancements include learned low-level bimanual manipulation, where neural networks translate visual inputs into precise physical actions. This synergy between pre-trained models for high-level planning and visuomotor policies for detailed execution enables the robot to handle complex tasks, like manipulating objects, with impressive efficiency and adaptability. The robot's ability to interpret context from conversations and visuals, combined with a robust memory system, allows it to execute tasks with a nuanced understanding of its environment and the commands given.

Visual-based newsletter on business and tech

We explain the latest business, finance, and tech news with visuals and data.📊

All in one free newsletter that takes < 5 minutes to read. 🗞

Save time and become more informed today.👇

Subscribe with one click today

World’s first major act to regulate AI passed by European lawmakers

The European Union's parliament has passed a groundbreaking regulatory framework for artificial intelligence, known as the EU AI Act, which categorizes AI technologies based on their risk levels, from unacceptable to low hazard. This legislation, set to be enforced starting May, aims to guide the development of AI while ensuring it aligns with human oversight and societal values. The act represents a significant stride in establishing global norms for AI governance, reflecting the EU's proactive stance in technology regulation, much like its earlier General Data Protection Regulation (GDPR).

The EU AI Act has garnered substantial support within the European Parliament, with a notable majority voting in favor. It's designed to foster innovation while protecting fundamental rights, highlighting the EU's ambition to set a global standard in AI regulation. However, the act's implementation remains a challenge, with concerns about its adaptability to the rapidly evolving AI landscape and the need for businesses to closely engage with the new regulations.

Critics and supporters alike acknowledge the AI Act's potential impact on the global stage, suggesting it could inspire similar legislative efforts in other countries. The act's introduction coincides with broader EU initiatives to address the dominance of major tech firms and the ethical implications of AI advancements. Legal experts and policymakers are now focused on translating the AI Act's provisions into effective practices, ensuring that the legislation evolves alongside the technology it seeks to govern.

Google’s new AI, SIMA, will play video games with you — but not to win

Google DeepMind has introduced SIMA, an AI training to play video games in a more human-like manner, currently in the research phase. SIMA, which stands for Scalable, Instructable, Multiworld Agent, aims to understand and play any video game, including open-world titles, through a blend of natural language instructions and image recognition. It's designed not to replace game AI but to integrate as a cooperative player, enhancing the gaming experience by learning from human interaction and game playthroughs.

Collaborating with 8 game developers like Hello Games and Coffee Stain, Google DeepMind has tested SIMA in various games, including No Man’s Sky and Goat Simulator 3, without needing custom APIs or game source codes. The training process involves a new Unity engine environment and recorded human-player interactions to teach SIMA basic skills and predict in-game outcomes, aiming to imbue the AI with a sense of spontaneity and strategic play.

While SIMA is still developing its repertoire of skills and adapting to new games, its future potential includes complex task execution and possibly communication within games. Unlike AI NPCs designed by companies like Nvidia, SIMA is envisioned as a dynamic player companion, contributing significantly to gameplay. Its progression could revolutionize AI's role in gaming, making it a versatile and engaging party member in various gaming environments.

Other stuff

• Security Flaws within ChatGPT Ecosystem Allowed Access to Accounts On Third-Party Websites and Sensitive Data

• AI reveals prostate cancer is not just one disease

• Another licensing Deal for OpenAI. This time with European Publishers Le Monde, Prisa

• Truffle-1 is an AI inference engine designed to run open-source models at home, on 60 Watts.

• Experts alarmed over AI in military as Gaza turns into “testing ground” for US-made war robots

• AI startup Cerebras unveils the WSE-3, the largest chip yet for generative AI

• Adobe Firefly repeats the same AI blunders as Google Gemini

• Amazon will let sellers paste a link so AI can make a product page

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.