✨ OpenAI's o3 breakthrough: nearly AGI

Also: New physics sim trains robots 430,000 times faster

Author

Evolving AI
December 23, 2024

In partnership with

Welcome, AI enthusiasts

OpenAI's newly announced o3 model – which is scheduled for release in early 2025 – is their new breakthrough reasoning model that shows that there’s no wall in AI. Genesis is a new physics engine that combines ultra-fast simulation with generative capabilities to create dynamic 4D worlds for robotics and AI training. And lastly, Apptronik partners with Google Deepmind to combine AI with robotics hardware, enabling humanoids to be more helpful to people in dynamic environments. Let’s dive in! 

In today’s insights:

  • OpenAI introduces o3, its most advanced AI model yet

  • New physics sim trains robots 430,000 times faster

  • Apptronik partners with Google DeepMind to advance humanoid robots with AI

Read time: 4 minutes

🗞️ LATEST DEVELOPMENTS

Source: OpenAI

Evolving AI: OpenAI announces o3, its breakthrough reasoning model that takes us close to AGI.

Key Points:

  • o3 gets record-high scores on important AI tests, showing great reasoning skills.

  • A smaller, cheaper o3 mini version is coming in January 2025 and is faster than the o1 model.

  • o3 solves problems by creating new solutions instead of just following patterns.

Details:

OpenAI’s o3 model is a major advancement in AI. It scored 75.7% on the ARC-AGI test, a key benchmark for measuring an AI model’s ability to handle complex math and logic problems it has never seen before. With extra computing power, o3 reached 87.5%, becoming the first model to surpass the human average threshold of 85% on ARC-AGI.

On the challenging Frontier Math test, o3 outperformed expectations. Math expert Terry Tao had predicted that these tests would resist AI for years, calling them "extremely challenging." Yet, o3 achieved a 25% success rate, far ahead of previous models that only managed 2%.

In coding competitions, o3 earned a Codeforces score of 2727, placing it 175th globally. This puts it in the top 0.01% of participants on the platform, which is already known for its highly skilled users.

Unlike earlier models, o3 doesn’t just rely on stored patterns or answers — it creates new programs to tackle difficult tasks. However, this advanced reasoning comes at a cost, requiring significant computing power and making it expensive to operate.

Why It Matters:

François Chollet, the creator of the ARC benchmark, describes o3 as a major departure from earlier language models. Unlike traditional pattern-matching approaches, o3 generates new programs on the fly to solve unfamiliar challenges. Still, Chollet points out that while impressive, o3 is not artificial general intelligence (AGI) and operates very differently from human cognition.

The company plans to release a more affordable o3 mini in late January 2025. Even at medium performance levels, this smaller model surpasses the capabilities of the earlier o1 system. The full version of o3 is expected to follow later.

Writer RAG tool: build production-ready RAG apps in minutes

RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.

Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.

Evolving AI: Scientists made Genesis, a tool that helps robots train way faster using super-speed simulations.

Key Points:

  • Genesis lets robots practice tasks 430,000 times faster in a virtual world than in real life.

  • It creates 3D training environments from text instructions and runs simulations really fast.

  • The system is open-source and uses Python, so lots of people can use it without expensive tools.

Details:

Genesis, a groundbreaking simulation platform, accelerates robotic training by compressing decades of learning into hours. Developed by a team led by Zhou Xian at Carnegie Mellon University, it outpaces existing simulators by processing physics 80 times faster. Using standard hardware, Genesis can run up to 100,000 simulations simultaneously. The platform also aims to integrate AI for generating dynamic, physics-rich virtual environments from text commands, potentially transforming how robots are trained and tested. The open-source system prioritizes accessibility, allowing researchers worldwide to leverage high-speed simulations for free.

Why It Matters:

The Genesis Project could change the future of robotics. By giving more people access to advanced training tools, it helps both experts and beginners come up with new ideas. Because it’s fast and affordable, Genesis could help robots become useful in areas like farming, building, and disaster recovery, making these jobs safer and more efficient.

Source: Apptronik

Evolving AI: Apptronik and Google DeepMind are teaming up to bring smarter humanoid robots into dynamic environments.

Key Points:

  • Apptronik aims to make humanoid robots practical for industries, starting with manufacturing and logistics.

  • Google DeepMind contributes advanced AI systems, enhancing robots' ability to reason and act in real-world settings.

  • Apollo, Apptronik’s flagship robot, is already undergoing trials with industry leaders.

Details:

Apptronik has announced a strategic partnership with Google DeepMind to combine AI with its humanoid robots. The Apollo humanoid robot, designed for physically demanding tasks, is a key focus. Apptronik’s team, with roots in NASA robotics projects, aims to improve industrial workflows by deploying these robots in manufacturing and logistics. Google DeepMind’s robotics team will provide AI expertise, building on its progress with models like Gemini and platforms like ALOHA. Apollo is already being tested by companies like Mercedes-Benz and GXO Logistics, highlighting its potential in industries requiring safe, collaborative robotics.

Why It Matters:

Apptronik is a robotics company focused on developing AI-powered humanoid robots designed to assist humanity in various tasks. The partnership with Google DeepMind marks Google’s return to the humanoid robotics field. This collaboration is likely to lead to significant advancements in the industry and more developments in the near future.

 👀 Click on the image you think is real

🔷 Sriram Krishnan named Trump’s senior policy advisor for AI.

🛡️ OpenAI trained o1 and o3 to ‘think’ about its safety policy.

🤖 Tetsuwan Scientific is making robotic AI scientists that can run experiments on their own.

🚗 MIT’s massive database of 8,000 new AI-generated EV designs could shape how the future of cars look.

📢 Instagram tests new AI-powered ad format for creators.

📞 Kalamazoo, MI, using AI to respond to non-emergency calls.

🛡️ AI cameras are giving DC's air defense a major upgrade.

🎥 TCL’s new AI short films range from bad comedy to existential horror.

📈 Trending AI Tools

  • 🎥 Kling AI v1.6 - Enhanced AI video generator with better prompts and pro modes (link)

  • 🧠 Gemini 2.0 Flash Thinking - Google DeepMind's new reasoning model, free to try (link)

  • 🚀 Hero - Sell stuff faster using AI (link)

  • ✍️ Trinka AI - AI Writing and Grammar Checker tool (link)

  • 💸 Expense Sorted - AI-powered tool that automatically categorizes your expenses (link)

  • 🤖 Tiledesk - Automate your customer service using AI agents (link)

  • 🖥️ Corexta - All-in-one business management platform (link)

Reply

or to participate.