DMflow.chat
ad
All-in-one DMflow.chat: Supports multi-platform integration, persistent memory, and flexible customizable fields. Connect databases and forms without extra development, plus interactive web pages and API data export, all in one step!
OpenAI’s 12-Day Major Update Breakdown: O1 Full Release, ChatGPT Pro, Sora Video Generation, O3 Overview
Introduction
OpenAI has released 12 major updates, including the full release of the O1 model, the ChatGPT Pro subscription plan, the revolutionary Sora video generation, and the new O3 model. Dive deep into the latest AI breakthroughs, from multimodal upgrades to superhuman intelligence performance.
OpenAI Day 1 Highlights: O1 Full Model & ChatGPT Pro
- O1 Full Model:
- Replaces the preview version, outperforming GPT-4 in math and coding, with faster speed and lower error rates.
- Adds multimodal input, capable of understanding images and text.
- General members: 20 uses per day
- ChatGPT Pro:
- For heavy users, priced at $200 per month.
- Unlimited access to O1, GPT-4, and advanced voice modes.
- Offers O1 Pro mode for handling more challenging problems.
Image Source: https://techwiser.com/day-1-of-12-days-of-openai-chatgpt-pro-and-o1-full-model-launched/
- The O1 model shows significant improvements over its preview version in competition math and coding, with accuracy increased by about 50%. This indicates a major boost in logical reasoning and programming capabilities.
- In doctoral-level science problems, the O1 model slightly underperforms compared to the preview version but still outperforms GPT-4 and human experts, showing its strength in handling complex issues.
-
This chart highlights the performance improvements of the O1 model over previous versions, especially in tasks requiring logical reasoning and programming.
- OpenAI Day 1: Launch of ChatGPT Pro at $200 per month, O1 full model available for paid users
OpenAI Day 2 Highlights: Enhanced Fine-Tuning (RFT)
- RFT:
- A new method to customize the O1-mini model using reinforcement learning and custom datasets.
- Enhances model reasoning in specific fields like law and medicine.
- Effective training with just a few examples.
-
Important for research, others can ignore
- OpenAI Day 2: Reinforcement Learning Fine-Tuning and Model Customization: The Future of AI
OpenAI Day 3 Highlights: Sora Video Generation Model
- Sora: Generates new videos from text, image, and video inputs, up to 1080p and 20 seconds long. Based on DALL·E and GPT models.
-
Usage: Available to ChatGPT Plus and Pro users, with Pro plan offering more usage and higher resolution. * $20 per month for 50 uses, cannot generate human faces * $200 per month for unlimited slow generation And provides three tools:
- Remix combines multiple video clips into a new video.
- Blend merges multiple videos for smooth transitions.
-
Loop creates a seamless loop from a single video clip.
- OpenAI Day 3: Leading Innovation! Sora Product Launch Event Recap
OpenAI Day 4 Highlights: Canvas Feature
- Canvas: Available to all users, integrated into the main model.
- Functionality: Allows running Python code and displaying results within Canvas, facilitating data analysis and visualization.
Not as good as Claude’s Artifacts or Cursor’s programming, but it has all the necessary features.
OpenAI Day 5 Highlights: ChatGPT x Apple Intelligence
- Siri will hand off tasks it can’t handle to ChatGPT
- Use ChatGPT to write documents
- iPhone 16 camera features
OpenAI Day 6 Highlights: ChatGPT Real-Time Interaction Upgrade
- Real-Time Interaction: Adds video functionality, screen sharing, and advanced voice modes.
-
Holiday Surprise: Introduces “Santa Mode.”
- OpenAI Day 6: Major Chatbot Upgrade: Real-Time Interaction and Holiday Surprises for a New Experience
OpenAI Day 7 Highlights: “Projects” Feature
- Projects: Integrates conversations with work scenarios, allowing file uploads, custom command settings, and project data organization.
- Applications: File organization, task automation, team collaboration.
Fills in the gaps left by Claude’s features.
OpenAI Day 8 Highlights: ChatGPT Search Feature
- Search Feature: ChatGPT can directly search the web for real-time information, available to all users globally.
- Highlights: Faster, more intuitive, and richer search experience, optimized for mobile and integrated with voice search.
Similar to the features provided by Perplexity’s Pro.
OpenAI Day 9 Highlights: Honoring Developers, Enhancing Developer Experience
- Highlights: Focuses on developers and startups, introducing new models and features to thank contributors.
- Main Updates:
- O1 Full Model: Ends preview, improves function calling, structured output, developer messages, inference effort parameters, and visual input.
- Real-Time API Updates: Supports WebRTC, reduces audio token costs (60% price drop for GPT-4), adds Python SDK.
- New Fine-Tuning Method: Introduces preference fine-tuning, optimizing model output with paired responses.
- OpenAI Day 9: Honoring Global Developers: Enhancing Developer Experience
OpenAI Day 10 Highlights: Full Integration with Phone, WhatsApp
- Highlights: ChatGPT is no longer limited to web and app, now accessible via phone and WhatsApp, lowering the barrier to AI use.
- Features:
- Phone Service: Call 1-800-CHAT-GPT (1-800-242-8478) to talk to AI.
- WhatsApp Integration: Interact with ChatGPT via WhatsApp, e.g., get recipes.
A feature for respecting the elderly.
OpenAI Day 11: Major Breakthrough in ChatGPT Desktop App: Comprehensive Analysis of New Generation AI Assistant Features
- Highlights: ChatGPT desktop app evolves from simple Q&A to a powerful smart assistant, offering cross-app integration, voice interaction, and programming support.
- Main Features:
- System Integration: Lightweight design, shortcut support, cross-app interaction.
- Programming Support: Supports mainstream IDEs, real-time code suggestions, Git command assistance.
- Document Processing Enhancement: Supports mainstream writing tools, smart content optimization, real-time document editing.
- Voice Interaction: Voice commands, personalized voice assistant, real-time responses.
ChatGPT x Mac
OpenAI Day 12: Major Breakthrough in AI Models: O3 Series Demonstrates Reasoning Beyond Human Capabilities
- Highlights: The O3 series models show exceptional performance in math and programming, surpassing humans in some tests.
- Major Breakthroughs:
- Programming: SWEET Bench Verified accuracy increased by 20%, Codeforces platform ELO scores significantly surpass O1.
- Math Operations: High scores on AMO benchmark tests, GPQ Diamond tests surpass O1 by 10 percentage points.
- O3 Mini: Lightweight model, cost-effective, outperforms O1 mini in multiple tests.
- Innovative Benchmark Tests: Breakthrough in Epic AI’s frontier math benchmark tests, ARC AGI benchmark tests first to surpass human average performance.
- Safety: Implements strict safety tests, adopts innovative safety training methods.
- Safety Testing Application Deadline: January 10, 2025
- O3-mini: End of January next year
-
O3 after O3-mini
- OpenAI Day 12: Major Breakthrough in AI Models: O3 Series Demonstrates Reasoning Beyond Human Capabilities
