Hub AI in practice
Artificial Intelligence is present in everyday life – from “googling” to facial recognition to vacuum cleaner robots. AI tools are becoming more and more elaborate and support people and companies more effectively in their tasks, such as generating graphics, texting or coding, or interpreting large amounts of data.
What AI tools are there, how do they work, how do they help in our everyday world – and how do they change our lives? These are the questions we address in our Content Hub Artificial Intelligence in Practice.
Zyphra has released Zonos-v0.1, an open source model that turns text into natural-sounding speech and can clone voices using just seconds of audio data. The new model supports five languages - English, Japanese, Chinese, French, and German - and gives users control over speaking speed, pitch, audio quality, and emotional tone. According to Zyphra, the model processes audio faster than real-time when running on an RTX 4090 GPU. Zyphra has made Zonos available in two versions: a pure transformer model and a hybrid model that combines state-space models with transformers. Both versions were trained on approximately 200,000 hours of audio data, primarily in English. Users can try out Zonos through a user-friendly Gradio interface, with easy Docker installation for local use. The model is also accessible through the Zyphra Playground or via API for those who prefer cloud-based solutions.
Today, we're excited to announce a beta release of Zonos, a highly expressive TTS model with high fidelity voice cloning.
We release both transformer and SSM-hybrid models under an Apache 2.0 license.
Zonos performs well vs leading TTS providers in quality and expressiveness. pic.twitter.com/jaliZNJecm
- Zyphra (@ZyphraAI) February 10, 2025
If you want to understand AI, there's one video you need to watch this week. Andrej Karpathy, formerly of OpenAI and Tesla, has released what might be the clearest explanation yet of how Large Language Models actually work. The video breaks down the entire training process of these AI systems and provides mental models for understanding their "psychology" - essentially, how they process and respond to information. Karpathy, who recently co-founded the AI education company Eureka Labs, also includes practical tips for getting the most out of these tools in real-world applications. What makes this explanation special is how Karpathy brings technical depth without sacrificing clarity.
Pika Labs' new "Pikadditions" tool allows users to seamlessly insert objects into existing videos. The process is simple: upload a video and an image, and the system places the object from the image into the video, trying to make it look as natural as possible. Users can create playful scenarios such as placing animals in washing machines or adding objects to videos of people opening refrigerators. While these may sound like simple effects, they represent a practical step forward in AI video editing. Pika Labs has recently focused on developing these easy-to-use AI video effects lately, rather than perfecting the ability to create entire videos from scratch. Their approach to "video inpainting" - while still a work in progress - can produce compelling results when the conditions are right.
OpenAI expands its European presence with its first German office in Munich. The company plans to build a local team focused on sales, development, communications and policy work, while strengthening its ties with German universities. According to CEO Sam Altman, Germany's reputation for technical excellence and industrial innovation made it a natural choice for expansion. The move is aimed at helping more German individuals, companies, and organizations make effective use of AI technology. Germany has become one of OpenAI's key markets globally, with ChatGPT usage showing particular growth: The company reports that its German user base has tripled in the past year, making it one of the platform's largest markets worldwide.