From December 5 to 20, OpenAI is organizing one conference per day, each time with one or more new features. This article summarizes the 12 announcements from the creator of OpenAI, if you don’t have time to follow the meeting daily.
To celebrate ChatGPT’s two years, OpenAI did not have a big press conference, as rumors suggested, but 12 mini-events. The artificial intelligence giant, which likes to be noticed, has launched an operation called “12 days of OpenAI”, which is nothing less than an advent calendar. Every day at 7 p.m. (excluding weekends), from December 5 to 20, it organizes a 10 to 15 minute live-stream from its offices in San Francisco, each time with one or more new releases. Google responded with a major update to Gemini, while xAI made Grok free.
In this article, Numerama offers you a brief recap of the 12 OpenAI conferences, in case you don’t have time to tune in every evening to listen to Sam Altman and his friends. It will be updated every evening until December 20. (A new image generation model is still expected as well as GPT-4.5, a major evolution of GPT-4).
Day 7: folders to organize your conversations in ChatGPT
On Friday, December 13, OpenAI revealed the arrival of a new “Projects” section on the ChatGPT home page. These are generally simple files, which allow you to store important conversations, in order to find them later (planning a trip or work in an apartment for example). Each project can have common files, as well as shared context elements between conversations.
The feature is already available with paid subscriptions and will arrive later in the free ChatGPT.
Day 6: a camera mode in ChatGPT Voice and Santa Claus
On Thursday, December 12, OpenAI unveiled several new features for the “advanced voice” mode of ChatGPT Voice, the service it launched in October 2024 with an ultra-realistic voice capable of imitating emotions.
The most impressive new feature is the addition of a video mode, which allows you to talk to ChatGPT with the camera on. The assistant takes regular screenshots of the feed, analyzes the content in the image and can answer specific questions like “am I dressed well”, “how to fix this piece of furniture” or “is this that this dish is cooked”, all without leaving voice mode.
Another new feature for the voice mode: a Santa Claus mode, which has special instructions. He answers all questions by vocally imitating the boss of the North Pole and brings all your requests back to the good Christmas mood. A real revolution for artificial intelligence!
Jour 5 : Apple Intelligence et ChatGPT
The announcement of Wednesday, December 11 is not one, since OpenAI took advantage of the public release of iOS 18.2 to reconsider its partnership with Apple. The company showed how an iPhone with Apple Intelligence enabled could now ask ChatGPT questions with Siri, simply summarize documents, generate text or analyze an image.
Day 4: Canvas, OpenAI’s Google Docs
Tuesday, December 10, OpenAI unveiled an update to Canvas, its interface dedicated to writing code or long text. This mode, which is accessible from the ChatGPT home page, aims to make the daily lives of people who write with generative AI easier. The generated text appears on the right of the screen, the discussion with the AI remains on the left. The idea is to use ChatGPT as an assistant to modify text or to have corrections suggested, without having to copy and paste everything each time.
Day 3: Sora, the tool for generating videos has finally arrived
For the moment, the announcement of Monday December 9 is, by far, the most spectacular. OpenAI has finally launched Sora, its tool for generating realistic videos, ten months after its first announcement. Sora is available in more than 100 countries, but not yet those in the European Union. Fortunately, it is possible to cheat with a VPN to try it in France.
According to our initial tests, Sora is excellent at generating simple things, but falters as soon as there is too much movement. The tool is nevertheless impressive, since it can turn any sentence into a video in just a few seconds. It is not the best in its category (on the open source side, there are several very efficient models that need to be run locally), but could emerge as a very popular solution in the future , especially to animate a still image.
Another important limitation: Sora requires a paid subscription. Most options are exclusive to the ChatGPT Pro subscription, with basic services in the ChatGPT Plus subscription.
This content is blocked because you have not accepted cookies and other trackers. This content is provided by YouTube.
To be able to view it, you must accept the use carried out by YouTube with your data which may be used for the following purposes: allowing you to view and share content with social media, promoting the development and improvement of products from Humanoid and its partners, display personalized advertisements to you in relation to your profile and activity, define a personalized advertising profile, measure the performance of advertisements and content on this site and measure the audience of this site (find out more more)
By clicking on “I accept all”, you consent to the aforementioned purposes for all cookies and other trackers placed by Humanoid and its partners.
You retain the option of withdrawing your consent at any time. For more information, we invite you to read our Cookies Policy.
I accept everything
Manage my choices
Day 2: fine-tuning, an announcement for developers
On Friday, December 6, OpenAI made arguably the least public announcement of its advent calendar. The company launched a research program on “Reinforcement Fine-Tuning,” a technique for customizing AI models to excel at complex, domain-specific tasks. Participants have access to an experimental API to train the models on their own data, evaluating answers and boosting performance on similar problems. This program is aimed at research institutes, universities and companies in sectors such as law, health or finance.
Day 1: o1 and ChatGPT Pro
On Thursday, December 5, OpenAI opened the first box of its advent calendar. Its first novelty is the arrival of the final version of the o1 model, which is an LLM capable of reasoning. By this OpenAI means that o1 is capable of asking questions based on your question, in order to propose a more complex answer, which involves solving several problems. o1 is available in ChatGPT.
In addition to the o1 model, OpenAI unveiled ChatGPT Pro, a new paid subscription at $200 per month. At this price there are no limits in the questions, as well as an even more sophisticated version of o1. ChatGPT Pro also integrates more options into Sora (announced in day 3) and could be essential for the final week’s announcements.
Related News :