OpenAI’s ChatGPT o3 - Pushing Humanity Forward!

Key Concepts

• OpenAI's o3: A new AI model with advanced image understanding, memory, and reasoning capabilities.
• Thinking with Images: The AI's ability to process and understand information from images, not just text.
• Humanity's Last Exam: A benchmark test for AI systems, consisting of challenging questions from scientists.
• Codex: A coding agent that allows users to create applications by referencing examples.
• Peer-Reviewed Sources: Scholarly articles that have been reviewed by experts in the field.
• System Prompt Customization: Tailoring the behavior of a chatbot by providing specific instructions.
• Convex Hull of Knowledge: The boundary of what an AI can do based on existing human knowledge.

1. OpenAI's o3: A Breakthrough in AI

• OpenAI showcased their new AI model, o3, which demonstrates significant advancements in thinking with images, learning, and potential job assistance.
• The AI's capabilities have drastically improved, with claims of a "genius level IQ," although the presenter expresses skepticism due to the lack of peer-reviewed publication.
• o3 can analyze images, identify objects, read obscured text, and provide information about the content.
  • Example: Identifying the largest ship in an image and its destination.
  • Example: Reading a tiny sign on a building after zooming in and cleaning up the image.
  • Example: Identifying movies filmed at a specific location.
  • Example: Finding Waldo in a picture.
• The AI can also mark up images to highlight specific details, such as issues with fabric samples.

2. Enhanced Memory and Personalized Learning

• o3 possesses improved memory capabilities, allowing it to recall past interactions and learn user preferences.
• The AI can use this information to teach users new things based on their existing knowledge and interests.
  • Example: Knowing the user likes scuba diving and music, o3 teaches them about coral larvae and their attraction to healthy reef sounds.
  • Experiment: Coral larvae are attracted to the sound of healthy reefs, leading to the idea of using underwater loudspeakers to attract them.
• The AI can also assist with job preparation by identifying knowledge gaps and providing relevant information.

3. Performance on "Humanity's Last Exam"

• "Humanity's Last Exam" is a challenging test designed to assess the capabilities of AI systems.
• o3 achieved a score of nearly 25% on the exam, a significant improvement from the previous best of 8% just a couple of months prior.
• This progress suggests that AI systems are becoming more capable of answering complex questions and potentially conducting research.

4. AI-Assisted Research and Innovation

• o3 can analyze incomplete research work and suggest ways to complete it, demonstrating its potential for assisting in scientific discovery.
• The AI can "think with images" during the research process, zooming in on relevant details and showing its reasoning.
• This capability suggests that AI systems are moving beyond simply working within the "convex hull of our knowledge" and may be able to push the boundaries of human understanding.
• The presenter hopes to see AI systems invent new things that humans have not yet discovered, leading to advancements in fields like drug design, agriculture, clean energy, and longevity.

5. Codex: Coding for Everyone

• Codex is a coding agent that simplifies the process of creating applications.
• Users can reference examples from social media or other sources to guide the AI in generating code.
• Example: Creating a real-time app that generates ASCII character art from an image using a camera.

6. Customizing Chatbot Behavior

• The presenter shares a trick for customizing chatbot behavior by providing specific instructions in the system prompt.
• By instructing the chatbot to "Look for peer-reviewed sources for every answer. Rate the credibility of that source on a scale of 0 to 10 for every source," users can improve the quality and reliability of the information they receive.
• This approach can be applied to any chatbot and allows users to refine the system prompt to meet their specific needs.

7. Conclusion

• OpenAI's o3 represents a significant advancement in AI capabilities, with improvements in image understanding, memory, reasoning, and research assistance.
• The AI's performance on challenging benchmarks like "Humanity's Last Exam" suggests that it is becoming more capable of answering complex questions and potentially contributing to scientific discovery.
• Tools like Codex and system prompt customization empower users to leverage AI for a wide range of tasks, from coding to research.
• The presenter emphasizes the importance of thorough research and careful evaluation of sources, encouraging viewers to engage in "Scholarly experiments" and share their findings.