The Innovations of OPENAI in Artificial Intelligence Technology (Part Two)

The Innovations of OpenAI: A Daily Journey Through the Future of Artificial Intelligence (Days 7 to 12)

As we previewed in yesterday’s post, OpenAI has revolutionized the artificial intelligence landscape and wrapped up the year with an impressive series of announcements during its "12 Days of OpenAI," showcasing significant advancements in AI technology. This article summarizes the most notable updates from Days 7 to 9 so that you don't miss any detail of this technological revolution.

Day 7: Projects in ChatGPT

OpenAI launched Projects in ChatGPT, a feature that allows users to organize conversations, upload files, and set custom instructions for each project. This tool facilitates the management of multiple tasks within the platform.

You can create projects, assign them a color, add files, and use custom instructions to focus the conversation. Projects work as "smart folders" to organize conversations. This feature enables file uploads, the use of search tools, and the Canvas feature. Users can also drag and drop chats into projects and utilize the conversation search tool.

Day 8: Search in ChatGPT for Everyone

ChatGPT’s real-time search feature, previously exclusive to paid users, is now available for all account holders for free. This improvement allows users to access updated web information directly from the chat interface, automatically determining if a query requires updated information or by using the web search button.

Additionally, the search experience has been optimized for mobile devices, offering enriched views of results. ChatGPT can also now be used as a default search engine in browsers, providing faster access to websites. The search integration has been extended to the voice mode, enabling web access via voice commands. Users can ask about events, weather, and more.

Day 9: New Tools for Developers

OpenAI introduced several updates for developers, including the general availability of O1 in the API with enhanced features like function calling, structured outputs, and developer messages. Developer messages are a new type of system message controlled by developers to guide the model's behavior.

Also introduced was deliberative reasoning, which allows adjusting the time the model spends "thinking" to save time and money. Vision inputs were added to the API, enabling models to process images with practical examples like detecting errors in forms. The model also demonstrated using functions to interact with APIs without user visibility and JSON output formatting. Internal evaluations showed that O1 surpasses GPT-4 in function calling and generating structured outputs.

On the topic of developer tools, I’ll soon share a post about Windsurf, a fantastic AI-powered application for programming, known for its generous free capabilities.

Day 10: ChatGPT on Your Phone

OpenAI expanded ChatGPT’s accessibility to phone calls and WhatsApp. In the U.S., users can call 1-800-CHATGPT to interact with AI via voice. Globally, users can engage with ChatGPT through WhatsApp.

This new functionality is designed for scenarios with limited data connectivity, allowing all users to leverage the power of AI. It can handle tasks like language assistance or explaining concepts. It has also shown its usefulness in providing cooking recipes, adapting in real time to user preferences, whether vegan or meat-inclusive.

Day 11: Working with Apps on Desktop

New functionalities were launched for ChatGPT desktop apps on Mac, enabling deeper interaction with computer applications. ChatGPT can now directly interact with content from other apps, such as terminals or code editors, automating tasks and analyzing information.

For example, ChatGPT can write commands in the terminal, generate graphs from results, or assist in writing code within an IDE. Support was also announced for new writing apps like Apple Notes, Notion, and Quip. Additionally, advanced voice mode with app-working functionality was added. This feature is available for the Mac desktop app and will soon be available for Windows.

Day 12: O3 and O3-mini Launch

OpenAI wrapped up its 12 Days of OpenAI with the announcement of two new models: O3 and O3-mini. O3 stands out for its advanced reasoning capabilities, demonstrating superior performance in benchmarks for coding, mathematics, and science. In the Epic AI mathematics benchmark, O3 achieved over 25% accuracy, whereas other models failed to surpass 2%.

Additionally, the model achieved a new result in the ARC AGI benchmark, with 87.5% accuracy, exceeding human performance. O3-mini, on the other hand, is a more cost-efficient model offering similar performance to previous models, with the added capability of adjusting reasoning time. Both models will undergo public safety trials before their general release. A new safety technique called deliberative alignment was also introduced, which uses reasoning to improve the detection of unsafe requests.

Conclusion

The "12 Days of OpenAI" have been a fascinating glimpse into the future of artificial intelligence, showcasing how this technology can transform our daily and professional lives. From organizing projects in ChatGPT to accessibility through phone calls and WhatsApp, OpenAI has demonstrated its commitment to innovation and inclusion.

The new tools for developers, the expansion of real-time search, and the introduction of advanced models like O3 and O3-mini underscore the limitless potential of AI to solve complex problems and enhance efficiency across various fields.

Finally, remember that if you’d like us to explore a specific topic, reach out to us at futuro@techglobalia.com. If you need specific information, it will be a challenge we’re excited to tackle.