Meeting Summary: 12 Days of OpenAI - Day 1

Attendees

• Jason
• Sam Altman
• Max
• Hyung

Transcript Summary

Sam Altman

• Introduction

  • Welcome to the 12 Days of OpenAI.
  • Daily launches or demos of new developments for the next 12 weekdays.
  • Aim to showcase recent work and provide a holiday present.

• First Launch: O1 Model

  • Full version of O1, an upgrade from O1 preview.
  • Enhanced intelligence, speed, multimodal capabilities, and better instruction following.
  • Significant improvements in math, competition coding, and GPQA diamond.
  • Demo to showcase performance in various tasks.

• Second Launch: ChatGPT Pro

  • New tier for power users needing more compute than the $20/month plan.
  • Unlimited access to models and advanced voice mode.
  • Introduction of O1 Pro mode for even better performance on complex problems.
  • Emphasis on reliability and significant performance boosts.
  • ChatGPT Pro priced at $200/month, launching today.

Hyung

• Introduction
  • Worked on building O1.
  • O1 model is faster and smaller than the O1 preview model.
  • Added multimodal input capability for reasoning through images and text.

Jason

• O1 Model Features
  • O1 thinks before responding, providing better and more detailed answers.
  • Rolled out to Plus and soon-to-be Pro subscribers on ChatGPT.
  • Faster and more intelligent responses compared to O1 preview.
  • Demonstrated with a history question, showing faster and more accurate responses.

Max

• ChatGPT Pro Features
  • Unlimited access to best models and advanced voice mode.
  • O1 Pro mode for enhanced performance on difficult problems.
  • Targeted at power users in programming and writing.
  • Demonstrated with a challenging chemistry problem, showing faster and more accurate results.

Hyung

• Multimodal Reasoning Demo
  • Demonstrated with a toy problem involving a data center in space.
  • Model handled ambiguity and provided accurate analysis.
  • Showcased state-of-the-art performance in benchmarks like MMM and math.

Max

• O1 Pro Mode Demo
  • Demonstrated with a challenging chemistry problem.
  • Model thought process and accuracy showcased.
  • Summary of O1's capabilities in reasoning over text and images, and solving complex problems.

Future Developments

• ChatGPT Pro Tier
  • More compute-intensive tasks and tools like web browsing and file uploads.
  • O1 model to be available on the API with new features for developers.

Closing Remarks

• Sam Altman
  • Congratulated the team on their work.
  • Preview of more developments for developers in the coming days.
  • Max shared a joke to conclude the session.

Max's Joke

• Joke
  • Santa was trying to get his large language model to do a math problem and he was prompting it really hard but it wasn't working. How did he eventually fix it?
  • Answer: He used reindeer enforcement learning.

This structured summary provides a clear and concise overview of the meeting, highlighting the key points discussed by each participant.