🧠 OpenAI unveils 'o3' & '03-mini' with near-AGI performance

Sign up | Sponsor

Hola, AI fam 🤖

In today’s WakeTheAI edition:

• OpenAI unveils ‘o3‘ & ‘o3-mini’

• Google DeepMind joins the humanoid race

• Prompt: Optimize Your Workflow With The Kanban Method

• Anthropic shares insights on building AI agents

• Google is reportedly adding an “AI Mode” to search

• Instagram teases AI tools to edit videos via prompts

• Google is expanding Gemini’s in-depth research mode to 40 languages

OPENAI
🧠 OpenAI unveils ‘o3’ & ‘o3-mini’

Lazy Bits: Yesterday, OpenAI unveiled “o3” and “o3-mini,” its next-generation reasoning models that push the boundaries of AI reasoning, achieving near-AGI performance under specific conditions.

In-Depth Details:

• Next-Level Reasoning: o3 models simulate human-like thinking by evaluating prompts, fact-checking outputs, and summarizing results—making them more reliable for complex tasks like coding, science, and math.

• Customizable Compute Levels: Users can set compute levels (low, medium, or high) to balance speed with accuracy, enabling deeper analysis for tasks requiring extended reasoning.

• Benchmark Dominance:

  • ARC-AGI Test: Achieved 87.5% on high compute—3x better than o1—showing progress toward AGI capabilities.

    

  • Programming Skills: Scored 2727 Codeforces rating (99.2nd percentile) and beat o1 by 22.8 points on SWE-Bench Verified.

    

  • Math Expertise: Solved 96.7% of the 2024 American Invitational Mathematics Exam, missing just 1 question.

    

  • Graduate-Level Science: Achieved 87.7% on GPQA Diamond for biology, physics, and chemistry questions.

  • Frontier Math Benchmark: Set a record at 25.2%, far exceeding competitors' 2% scores.

    

• AGI Potential: While o3 approaches AGI-level reasoning in some areas, it still struggles with simple tasks, highlighting gaps that separate it from human intelligence.

• Availability: o3-mini is available for safety researchers now, with public rollout expected in late January, followed by o3 later pending safety reviews.

Lazy Conclusion: OpenAI’s o3 models raise the bar for AI reasoning, hinting at AGI potential while outperforming rivals in benchmarks. With high-performance math, science, and programming abilities, they could reshape AI’s role in problem-solving, but gaps remain before true AGI is achieved.

TOGETHER WITH HUBSPOT

Ready to Level up your work with AI?

HubSpot’s free guide to using ChatGPT at work is your new cheat code to go from working hard to hardly working

HubSpot’s guide will teach you:

• How to prompt like a pro

• How to integrate AI in your personal workflow

• Over 100+ useful prompt ideas

All in order to help you unleash the power of AI for a more efficient, impactful professional life.

Get the free guide and level up your AI game today!

ROBOTICS
🤖 Google DeepMind joins the humanoid race

Lazy Bits: Apptronik has partnered with Google DeepMind to advance AI-powered humanoid robots like Apollo, focusing on manufacturing, logistics, and eldercare applications.

In-Depth Details:

• Humanoid Design: Apollo, at 5'8" and 160 pounds, is built for safety and reliability, designed to handle physically demanding tasks alongside humans.

• AI Integration: Google DeepMind’s AI enhances Apollo’s adaptability and dexterity for tasks like material sorting, heavy lifting, and eldercare assistance.

• Industrial Applications: Apollo targets warehouses, manufacturing, and logistics, addressing labor shortages by streamlining repetitive and complex tasks.

• Healthcare and Eldercare Focus: Apollo is set to assist with daily tasks for seniors, improving quality of life and reducing caregiver workloads.

• Strategic Partnerships: Apptronik is collaborating with Mercedes-Benz and GXO Logistics to test Apollo’s applications in manufacturing and warehousing environments.

Lazy Conclusion: Apptronik and Google DeepMind’s partnership could redefine robotics, solving labor shortages and eldercare needs while pushing AI-powered humanoids into mainstream applications.

Optimize Your Workflow With The Kanban Method

You can ask ChatGPT to act as your Kanban implementation expert, providing a detailed step-by-step guide to streamline workflows using the Kanban Method.

It will cover core principles, board setup, task management, and strategies for visualizing progress and resolving bottlenecks.

The output will be structured in a numbered format with actionable tips, making it easy to apply across industries and team sizes.

Prompt:

Act as a Kanban implementation expert specializing in workflow optimization. Your task is to create a detailed step-by-step guide for setting up and utilizing the Kanban Method to improve productivity. The output should address the following key areas:

1. Introduce the fundamental principles of Kanban and how they apply to workflow optimization.
2. Provide clear, actionable instructions for setting up a physical or digital Kanban board, including layout suggestions.
3. Outline methods for breaking down tasks into manageable cards and categorizing them effectively.
4. Explain the process of moving cards through workflow stages and managing progress visually.
5. Share best practices for fostering team collaboration, identifying bottlenecks, and promoting continuous improvement through regular reviews.

Key Information to Include:

Project/Process Focus: [Insert details about your specific project or process here]
Team Size: [Specify team size]
Industry Context: [Clarify industry for tailored strategies]
Current Workflow Issues: [Highlight any obstacles or inefficiencies in the current process]

Output Requirements:

• Present the guide in a numbered list format for easy navigation.
• Use bullet points for sub-points to provide detailed explanations and relevant examples where applicable.
• Ensure the language is practical, actionable, and adaptable to different industries and team sizes.

Result:

• Google is reportedly adding an "AI Mode" to Search, offering chatbot-style answers, voice queries, and follow-up prompts, aiming to rival ChatGPT's search features.

• Anthropic shares insights on building AI agents, favoring simple patterns over complex frameworks for reliability, scalability, and ease of debugging.

• Google is expanding Gemini’s in-depth research mode to 40 languages, offering AI-powered multi-step research and reports, with efforts to improve accuracy.

• Instagram head Adam Mosseri teases AI tools to edit videos via prompts—change outfits, backgrounds, and appearances—powered by Meta’s Movie Gen AI, launching next year.

1. 🤖 Faune: Have an anonymous AI chat with Dynamic LLMs.

2. 📊 Webscrape: Automate your data collection with no code.