12 Days of OpenAI - Comprehensive summary.
Comprehensive update on OpenAI's '12 Days of OpenAI' , ship-mas campaign, featuring groundbreaking updates like Sora, O3, Canvas, Projects, Agents with Mac and a lot more.
OpenAI has released a series of exciting products and updates during the "12 Days of OpenAI" campaign beginning 5th December 2024 and ending on 20th December 2024. Some updates have struck a deep chord with users, becoming instant favorites, while others may need more time to resonate. In this blog, I’ll summarize each release day by day, highlight key features, and share my personal favorites.
Key Themes of the Releases by Open AI
Wider Accessibility: OpenAI is focused on making ChatGPT more widely accessible. This includes launching 1-800-ChatGPT, a phone and WhatsApp-based service allowing interaction without an account.
Multimodal Expansion: A significant push towards multimodal capabilities is evident with the introduction of Sora, a text-to-video generation model, and the integration of live video and screen share features in ChatGPT's advanced voice mode.
Enhanced User Experience: Existing features like ChatGPT search are being improved with faster results, map outputs, and integration with voice conversations.
New Features and Functionality: Noteworthy additions include ChatGPT Projects for organising chats and files, Apple Intelligence integration, and Reinforcement Fine-tuning for developers.
Day wise summary of key features by OpenAI
Day 1 (December 5th): ChatGPT Pro
ChatGPT Pro Plan: This plan is designed for users who need enhanced AI capabilities. It provides unlimited access to o1, OpenAI's most powerful model, along with o1-mini, GPT-4o (for multimodal applications), and Advanced Voice features.
How it works: Users pay a monthly subscription fee of $200 to gain unrestricted access to these advanced models, enabling them to tackle more complex and demanding AI tasks. If you are not sure if this subscription fee is justified , do read my blog that I have written earlier -
Key features:
Unlimited o1 access: Unlocks the full potential of OpenAI's most advanced model for tackling intricate problems.
o1 Pro Mode: Utilises greater computational power for improved accuracy and comprehensiveness, especially in challenging fields like data science, programming, and legal analysis.
Uniqueness: Offers unprecedented access to cutting-edge AI capabilities in a user-friendly format, empowering users to leverage the latest advancements in AI.
Benefits:
Researchers and Engineers: Accelerate research, development, and analysis with access to powerful AI models.
Professionals in Data Science, Programming, and Law: Enhance productivity and decision-making with more accurate and insightful AI assistance.
ChatGPT Pro Grants: To support progress in areas that benefit humanity, OpenAI awarded ten grants of ChatGPT Pro to medical researchers at prestigious US institutions. These grants aim to further healthcare advancements by providing researchers with access to the most advanced AI tools.
Day 2 (December 6th): Reinforcement Fine-Tuning Expansion
Expansion of the Reinforcement Fine-Tuning Research Program: This program enables developers and machine learning engineers to create highly specialized models fine-tuned with specific data sets to excel at particular, complex tasks. This is different from standard fine-tuning and leverages reinforcement learning algorithms. This is the same technique used to train OpenAI's frontier models, GPT-4o and the o1 series.
How it works:
Users supply a data set of examples and define a 'grader' to assess the model's performance.
The model is trained using reinforcement learning, reinforcing correct responses and discouraging incorrect ones.
OpenAI takes care of the training once the user provides the data set and grader.
Key features:
Tailored to complex, domain-specific tasks: Creates expert models that excel in specific areas.
Enables the creation of unique AI offerings: Users can develop AI solutions for their specific needs and markets.
Benefits:
Developers and ML Engineers: Create bespoke models that outperform general-purpose models on specific tasks.
Businesses and Organisations: Develop unique AI solutions with competitive advantages.
Alpha Program: OpenAI is expanding its alpha program via the Reinforcement Fine-Tuning Research Program to enable more people to test the capabilities of o1 models. The program is best suited for organisations working on complex tasks with expert teams who believe AI assistance would be beneficial.
This program allows users to leverage their domain expertise and data sets while utilising OpenAI's reinforcement learning algorithms and model training infrastructure.
Day 3 (December 9th): Sora Video Generation
Launch of Sora, the Text-to-Video Generation Model: Sora generates realistic videos from text prompts, images, or video files. It is currently available to ChatGPT Plus, Team, and Pro users. Free, Enterprise, and Edu accounts are not eligible.
How it works: Users input text descriptions, images, or video segments, and Sora generates video content based on these inputs. You can access the model at sora.com. Sora Turbo is a new high-end accelerated version of the original Sora model.
Key features:
Sora Video Editor: Enables users to generate videos up to 20 seconds long while maintaining visual quality and adherence to the prompt. Users can change the aspect ratio, resolution, duration, and the number of variations to create. There are also presets that can be used, or users can create their own.
Editing options: Includes re-cut, remix, blend, and loop functions for further customisation. Learn more about the features and how it will shape the future of Movie making in my blog I wrote earlier.
Sharing and Management: Users can favourite, share, download, organise into folders, report, or delete videos. Deleted content cannot be recovered.
Storyboard feature: Allows users to create videos frame by frame, adding captions and defining the sequence. The timeline can be adjusted to determine the pace and allow for connecting scenes or cinematic cuts. Users can upload an image into a storyboard card to create a video.
Benefits:
Content Creators: Streamline video production and explore creative concepts. Visualise your stories quickly and modify them with ease.
Artists and Designers: Experiment with visual storytelling and animation.
Day 4 (December 10th): Canvas Enhancements
Canvas Becomes Default in GPT-4o: The interactive writing and coding interface, Canvas, is now the standard interface for GPT-4o, available to all users.
How it works:
Users can access a blank canvas through various methods, including requesting one from ChatGPT, pasting content, or using specific commands.
Canvas enables real-time collaboration with Chat GPT itself, editing, and feedback within a shared workspace. Read more about it in my previous blog:
Key features:
GPT integration: GPTs can now leverage Canvas for enhanced interactions.
Python code execution: Directly run Python code in Canvas with bug-fixing and error commenting assistance.
Uniqueness: Offers a seamless and intuitive environment for collaborative writing and coding tasks, leveraging GPT-4o's capabilities.
Benefits:
Writers and Coders: Improved writing and coding experience with collaborative features and AI assistance.
Teams: Enhanced collaboration and productivity.
Day 5 (December 11th): Apple Intelligence Integration
ChatGPT Integrated with Apple Intelligence: Seamless integration of ChatGPT within iOS, iPadOS, and macOS, enhancing user experience and productivity. This is a game changer for Apple users for sure. Your favourite Siri just got smarter and more capable.
How it works: ChatGPT becomes deeply embedded within Apple's operating systems, accessible through various system functions.
Key features:
Siri Handoff: Siri can delegate tasks to ChatGPT, streamlining user interactions.
Writing Tools Integration: ChatGPT empowers Apple's Writing Tools with advanced capabilities for composing, refining, and summarising text.
Visual Intelligence in Camera: iPhone 16 users can leverage ChatGPT to analyse and understand visual information through the camera app.
Uniqueness: Combines the power of ChatGPT with the user-friendly interface of Apple devices, making AI more accessible.
Benefits:
Apple Users: Enhanced productivity and creativity across various tasks.
Developers: Opportunities to create innovative applications leveraging this integration.
Day 6 (December 12th): Video in Advanced Voice & Santa Mode
Video and Screensharing in Advanced Voice Mode: Enables real-time video sharing and screensharing during voice conversations with ChatGPT.
How it works: Users can activate video and screensharing directly within the Advanced Voice Mode interface.
Key features:
Enhanced Visual Communication: Facilitates richer and more engaging interactions.
Improved Collaboration: Enables real-time visual collaboration and problem-solving.
Benefits:
Users: More dynamic and effective communication with ChatGPT.
Businesses: Potential for innovative applications in remote work, customer support, and training.
Santa Mode Launch: Allows users to engage with ChatGPT using a festive Santa voice, enhancing the holiday experience.
Day 7 (December 13th): Projects Feature
Introduction of Projects in ChatGPT: Offers a new way to organise and manage ongoing work or personal projects within ChatGPT, available to Plus, Team, and Pro users. This is very similar to Claude Projects , Perplexity Spaces and other tools .
How it works: Projects provide dedicated spaces where users can group relevant chats, files, and custom instructions.
Key features:
Centralised Workspace: Consolidates information and conversations related to a specific project.
Custom Instructions: Allows users to set specific guidelines and preferences for each project.
Integration with Other Features: Supports Canvas, Advanced Data Analysis, DALL-E, and Search within Projects.
Uniqueness: Enhances ChatGPT's capabilities for managing complex tasks and projects, improving organisation and workflow.
Benefits:
Users: Increased efficiency and focus when working on multiple projects or complex tasks.
Day 8 (December 16th): ChatGPT Search Enhancements
Improvements to ChatGPT Search: Enhances search functionality within ChatGPT, offering faster results, map outputs, and integration with voice conversations. It’s trying to give a tough competition to Google Search and Perplexity AI
How it works:
Search results are delivered more rapidly, and map outputs are now available on mobile devices.
Users can initiate searches using voice commands during voice conversations.
Key features:
Faster Search Results: Provides quicker access to information.
Map Outputs: Offers visual representations of location-based searches.
Voice Search Integration: Enables seamless search functionality during voice interactions.
Uniqueness: Improves the efficiency and convenience of using search within ChatGPT.
Benefits:
Users: Enhanced search experience with faster results and integrated voice capabilities.
Day 9 (December 17th): Developer-Focused Updates
New Tools and Upgrades for Developers: Introduces more capable models, customisation tools, and performance enhancements for developers utilising OpenAI's API.
How it works: These updates provide developers with advanced features and tools to build more sophisticated and efficient AI applications.
Key features:
OpenAI o1 in the API: Access to the powerful o1 model with enhanced capabilities like function calling, structured outputs, and vision processing.
Realtime API Updates: Improved voice quality, reduced pricing, and more control over responses for real-time conversational applications.
Preference Fine-Tuning: A new technique for customising models based on user and developer preferences.
New SDKs: Beta versions of Go and Java SDKs expand the range of supported programming languages.
Uniqueness: Equips developers with cutting-edge tools and models to create innovative and impactful AI solutions.
Benefits:
Developers: Enhanced capabilities and flexibility in building AI applications.
Businesses and Organisations: Access to a wider range of AI-powered solutions.
Day 10 (December 18th): 1-800-ChatGPT
Launch of 1-800-ChatGPT: Enables access to ChatGPT via phone calls and WhatsApp messaging without requiring an account. The number is 1-800-242-8478 .
How it works: Users can initiate conversations by calling a dedicated phone number or messaging via WhatsApp. Scan to QR Code below to chat on Whatsapp, if you haven’t done that yet.
Key features:
Phone Call and WhatsApp Access: Expands accessibility to users without accounts or internet access.
Free Usage: Offers 15 minutes of free usage per month.
Uniqueness: Makes ChatGPT more accessible to a broader audience, including those who may not have traditional internet access.
Benefits:
Users in Underserved Areas: Access to AI assistance and information via phone calls.
Businesses: Potential for new customer service and communication channels.
Day 11 (December 19th): macOS App Enhancements - making ChatGPT Agentic
Enhanced Functionality for ChatGPT on macOS: This update introduces new features and improves existing ones for a smoother user experience on macOS devices. These features are available on ChatGPT for macOS version 1.2024.346 or later.
How it works: These updates focus on making ChatGPT more integrated and user-friendly within the macOS ecosystem.
Key features:
"Work with Apps": Allows ChatGPT to access content from compatible apps, providing smarter and more tailored responses. This feature requires macOS Accessibility API permissions to function. ChatGPT will never look at the contents of another app unless the user explicitly selects the app.
Users can see what content ChatGPT will include before sending their message.
Content from apps is included in the chat history.
Advanced Voice Mode with App Integration: Users can use Advanced Voice Mode to interact with ChatGPT while working with other applications.
Conversation Search: Enables users to search through their past conversations using keywords and phrases by clicking on the search bar.
Expanded App Support: The update adds support for a wider range of note-taking and coding applications, including Apple Notes, Notion, Quip, Warp, Xcode, VS Code (including Code, Code Insiders, VSCodium, Cursor, and Windsurf), JetBrains IDEs (including Android Studio, IntelliJ, PyCharm, WebStorm, PHPStorm, CLion, Rider, RubyMine, AppCode, GoLand, and DataGrip), TextEdit, Terminal, iTerm, and Prompt.
Benefits:
macOS Users: Increased productivity and efficiency when using ChatGPT on macOS devices, alongside smoother integration with various apps and workflows.
This update represents a shift towards ChatGPT being more agentic, going beyond simple questions and answers, and allowing users to automate more tasks directly within their macOS environment.
Day 12 (December 20th): Announcing o3 and o3-mini . Steps towards AGI
Announcement of o3 and o3-mini reasoning models: o3 is a highly advanced reasoning model and o3-mini is a cost-efficient reasoning model.
How it works: o3 and o3-mini are advanced reasoning models trained to excel in complex tasks and benchmarks.
Key Features:
o3-mini performance and efficiency: o3-mini demonstrates significant improvements in performance and cost-efficiency compared to its predecessor, o1-mini, especially in coding and mathematics. It is also faster than o1.
Adaptive Thinking Time: o3-mini supports adaptive thinking time with options for low, medium, and high reasoning effort, allowing users to adjust processing time based on task complexity.
Support for API features: o3-mini supports function calling, structured outputs, and developer messages, providing developers with more versatile tools for building advanced applications.
o3 benchmark performance: o3 showcases exceptional performance in various technical benchmarks, particularly in coding, mathematics, and reasoning.
Coding Benchmarks: o3 excels in coding benchmarks, such as SWE-bench verified and HumanEval, demonstrating advanced code generation and problem-solving capabilities.
Mathematics Benchmarks: o3 achieves state-of-the-art results in mathematics benchmarks like MATH and AIME, showcasing its proficiency in complex mathematical reasoning and problem-solving.
Reasoning Benchmarks: o3 also achieves impressive results in reasoning benchmarks, such as ARC Challenge, signifying its advanced reasoning capabilities. The challenge is created by ARC Prize Foundation
Uniqueness: This release marks a significant step towards the next phase of AI, characterised by models capable of handling increasingly complex tasks that require advanced reasoning.
Benefits:
Researchers: Access to powerful models for pushing the boundaries of AI research in areas like coding, mathematics, and reasoning.
Developers: Increased potential for building advanced AI applications leveraging the enhanced capabilities of o3 and o3-mini.
Availability: While not publicly launched on December 20th, 2024, applications for safety testing were open from December 20th, 2024 to January 10th, 2025. OpenAI planned to launch o3-mini around the end of January 2025 and the full o3 shortly after that.
My Favorites
Out of all these releases, a few stood out for me:
Sora : Text to Video Model with advanced features
Launch of O3 & O3-mini : The most advanced models and stepping towards AGI
Canvas Feature : The best way to write content , code and collaborate with Chatgpt
Connect to Mac Apps : Get context about the work you are working and start solving it with complete context about the issue.
1-800-ChatGPT: The convenience of accessing AI via phone is next-level.