After the holiday rush of AI announcements in December, the first week of 2025 felt quieter—but not completely silent. Here’s a quick look at everything that happened in AI last week (December 29, 2024, to January 5, 2025). Let’s dive in.
Table of Contents
1. Meta Deleted AI Profiles After User Backlash
Meta rolled out 28 AI-generated users on Facebook and Instagram back in September 2023. Last week, Connor Hayes, VP of Generative AI at Meta, told the Financial Times that these AIs would eventually function like regular accounts—with bios, profile pictures, and AI-generated content. While most social media sites are figuring out how to block bots, Meta adding AI bots into its apps might be the most dystopian AI move so far.
As expected, things didn’t go as planned. Users found these AI profiles confusing, irrelevant, and even a bit creepy. Some couldn’t block them due to a bug, which only made things worse. After the backlash, Meta pulled the profiles and promised fixes this week.
HuggingFace Introduced Smolagents
HuggingFace unveiled Smolagents, a new tool for building lightweight AI agents. It lets open-source LLMs connect to services like Google Maps or search engines, making them act instead of simply responding. Developers can quickly build smart assistants for research, travel planning, or data analysis using simple Python code. Smolagents replaces Hugging Face’s older transformers agents library with a faster, more flexible solution.
TangoFlux– The Fastest Text-to-Audio Model Released
TangoFlux is the next text-to-audio AI model that can generate 30 seconds of 44.1kHz stereo audio in just 3.7 seconds. The tricky part with training audio models is that they don’t have clear right or wrong answers like text models. To fix this, TANGOFLUX uses a method called CLAP-Ranked Preference Optimization (CRPO) to fine-tune its outputs.
It already outperforms popular models like AudioLDM2 and AudioBox in both quality and detail. Additionally, the code and models are open-source, allowing anyone to try them out on the HuggingFace website.
OpenAI Plans to Become a for-Profit Company
OpenAI started as a non-profit in 2015 to ensure AI benefits everyone. As AI advanced, it needed more funding, so it added a for-profit arm in 2019 to raise billions for development. Now, to handle even bigger costs, OpenAI plans to turn its for-profit side into a Public Benefit Corporation (PBC). The non-profit arm will focus on charitable projects, while the for-profit side will raise capital and develop AI systems. This move helps OpenAI compete with rivals like Anthropic and xAI, which already use similar structures.
Elon Musk’s legal battle against OpenAI has gained backing from Geoffrey Hinton, often called the ‘godfather’ of AI. Hinton’s involvement adds weight to Musk’s claims that OpenAI has strayed from its founding principles.
OpenAI Failed to Launch Media Manager
OpenAI’s Media Manager tool, announced in May to help creators control how their work is used for AI training, is still missing despite promises to launch it before 2025. It was meant to let creators opt in or out of AI training, but reports suggest it was never a priority internally.
With OpenAI facing multiple copyright lawsuits, the tool was seen as a way to ease concerns and legal risks. However, delays and silence about its progress have raised doubts about whether it will launch anytime soon—or at all.
Nintendo Switch 2 Will Include AI Features
Leaks about the Nintendo Switch 2 suggest it could feature AI-powered upscaling using Nvidia’s DLSS technology. This would allow the console to upscale lower-resolution images to 4K, making games look sharper and more modern. If true, this will improve the gameplay experience—especially when docked to a TV.
While nothing is confirmed, AI-powered upscaling could make the Switch 2 a big step forward for handheld and console gaming.
Also Read:
- DeepSeek V3: Can Free and Open-Source AI Chatbot Beat ChatGPT and Gemini
- Which Gemini Model to Choose – 1.5 Flash, 1.5 Pro, Deep Research, 2.0 Flash or Advanced
- NotebookLM Gets a New Look and Lets You Join AI Podcast
- 12 Days of OpenAI: Everything Announced During the Campaign
Backflip – New AI Tool for Generating 3D Printable Objects
Backflip, a newly launched AI tool, allows users to generate 3D printable designs for free. Simply sign up and provide a prompt to generate images of the object. Then, convert them into 3D models with the click of a button. You can also download them afterward.
However, due to heavy demand, Backflip is limiting access to new users. However, you can log in today to join the waitlist.
3D Traj Master Can Control Motion in Generated Video
We have many video generation models like Sora and Veo2. While we can define action through the prompt, we cannot draw a trajectory to show how the entity should move. With 3D Traj Master, you can give a prompt like “dog walking on a beach” and also provide a 3D path showing how the dog should walk. It will generate a video where the dog walks along the specified path.
The platform also provides the option to change the subject, so you can remove the dog and add a car. You can also change the background. It is an open-source model, but currently, it does not offer a cloud-based version for direct use. The primary way to access its functionalities is by downloading and installing the software from its GitHub repository.
