AI at Meta’s Post

Very exciting that Meta is creating tech to watermark AI generated videos. This should be mandatory for all AI generated videos so bad actors can be spotted.

Thanks for sharing AI at Meta. Over the past decade interest in traceability and content IP has grown. At IBM we were able to trace digital content that was used by IBM Watson and then in HP invisibly watermark physical content produced through our printers. With the introduction of Meta Video Seal, being able to watermark video for artists and content providers is awesome! Still work to be done on LLMs and Foundation Models to be able to answer “where did you get that from?” and “why do you think that?”

Our team just released Video Seal, a state-of-the-art, open-source watermarking model for video, under a commercial license (MIT) As AI generated content becomes increasingly prominent online, and generative ai models more and more realistic, invisible watermarking emerges as a promising solution. Video Seal offers several key features: - fully open source, with a commercial license - High speed - SOTA robustness across a wide range of common video edits (e.g. cropping, flip, brightness changes) as well as video compression commonly used on online platforms Try out the demo today! Models & Training Code: https://lnkd.in/dawyiDE5 Read the Paper: https://lnkd.in/dmEmsZxM Congrats again to our team Pierre Fernandez Hady Elsahar I. Zeki Yalniz

👆 #AI #Video #watermarking plays a crucial role in distinguishing fake videos from authentic ones and in protecting #intellectual #property rights. ☑️ #Video #Seal #Meta #Neural #Fake #Innonvation #AI #Media #Production

Here's how Video Seal could be a valuable tool for engineers: 1. Protecting Designs and Intellectual Property: Watermarking CAD files and simulations: Engineers often create valuable designs and simulations using CAD software or other specialized tools. Video Seal can be used to embed watermarks in videos of these designs, making it difficult for others to steal or claim ownership of their work. This is particularly important in competitive industries or when sharing designs with clients or collaborators. Securing training and instructional videos: Engineers often create training videos or tutorials for internal use or to share knowledge with others. Watermarking these videos with Video Seal can help prevent unauthorized distribution or modification. 2. Ensuring Data Integrity and Authenticity: Verifying data from autonomous systems: Engineers working with autonomous systems (e.g., drones, robots) rely heavily on video data for navigation, object recognition, and decision-making. Video Seal can help verify the authenticity and integrity of this data, ensuring that it hasn't been tampered with or corrupted. This is crucial for safety and reliability in applications like self-driving cars and industrial automation. Authenticating video evidence in investigations: In fields like civil engineering or failure analysis, video footage can be crucial evidence. Video Seal can help verify that video evidence hasn't been manipulated, ensuring accurate investigations and reliable conclusions. 3. Building Trust and Transparency in AI-Generated Content: Verifying AI-generated designs and simulations: As AI plays a larger role in engineering design, it's important to be able to verify the outputs of AI models. Video Seal can be used to watermark AI-generated videos of designs or simulations, providing a way to track their origin and ensure they haven't been altered. Increasing confidence in AI-driven decisions: In applications where AI is used to make critical decisions (e.g., structural analysis, medical diagnosis), Video Seal can help build trust by providing a way to verify the integrity of the AI's input data and output visualizations. 4. Other Potential Benefits: Tracking the provenance of video data: Video Seal can help track the origin and history of video data, which can be useful for managing versions, identifying sources of errors, and ensuring compliance with regulations. Detecting deepfakes and misinformation: As deepfake technology becomes more sophisticated, Video Seal could help engineers identify manipulated videos that might be used to spread misinformation or cause harm. By understanding these potential benefits, engineers can leverage Video Seal to protect their work, ensure data integrity, and promote responsible use of AI in their respective fields.

I am excited to see where this leads us in the future regarding traceability, observability, and liability of digital content, especially with the proliferation of generative AI. This is just the first step in building a robust data supply chain and improving traceability. I am also curious about the potential effects on AI training if future datasets include these invisible watermarks. Will they impact the training data and model behaviour if we don't cleanse them first? Alternatively, will we be able to trace the origins of datasets through these watermarks, allowing for better management of ownership and copyright issues?

This is game changer from Meta. Imagine we use this same tech of watermarking at media outlets to distinguish between real and deepfake. More and more companies need to bring this particular tech to their platforms to combat deepfake. Maybe assigning unique blockchain IDs to VIPs and Politicians and using a water mark technique to flag their content as original once the keys match.

Just finished Day 11 of my #30DaysOfFL journey—this time exploring **Federated Learning for Text Generation**! 🚀 If you’re curious about how pre-trained language models (think character-level RNNs) can be fine-tuned on decentralized data (like lines from Shakespeare’s plays) *without* pooling all that text into one spot, this is for you. **What’s inside the repo?** - A walkthrough of taking a Dickens-inspired model and using Federated Averaging to tailor it to Shakespearean text. - Insights into handling non-IID (non-uniform) data distributions—common in real-world federated setups. - Steps to seamlessly push updated model weights back into Keras, making evaluation and text generation effortless. - Practical snippets that show exactly how to integrate TFF with your standard ML workflow. In short, you’ll learn how to:  - Adapt a pre-trained language model in a privacy-preserving way.  - Leverage the TFF framework for text data.  - Get hands-on with code that’s both academically intriguing and practically valuable. Check out the GitHub repo to dive deeper: https://lnkd.in/er7wKQD2 #FederatedLearning #MachineLearning #NLP #DataScience #30DaysOfFLCode

🔍 How can we push the boundaries of video LLMs for temporal reasoning? Our latest work introduces: ✔️ A time-aware clip encoder for capturing precise short-term temporal cues. ✔️ A BiLSTM module for understanding long-term dependencies in lengthy videos. This milestone wouldn’t have been possible without the brilliant work of my fellow authors, Umer Ahmed and Hamza Khan, and without the invaluable mentorship of Quoc-Huy Tran and Zeeshan Zia—thank you all for your dedication and guidance! 🙌 Check out our results in the link/video below and share your feedback! #GenerativeAI #LLMS #DeepLearning #MultimodalAI #Innovation

When you watch a long video and only get to see a few clips, it’s easy to miss important details and lose track of the overall sequence of events. Video LLMs suffer from this problem—sampling only a handful of frames from a long video—which means they often miss crucial context and can’t accurately process or describe what’s happening from start to finish. We've developed a new approach that harnesses the power of a Bidirectional LSTM to significantly enhance the temporal reasoning ability of video LLMs. By encoding time-aware clip features and aggregating them into a single global representation, we’ve achieved state-of-the-art results in tasks like dense video captioning, temporal grounding, highlight detection, and action segmentation. Check out our intro video for a quick overview and visit our arXiv link for the full details! Project Page: https://lnkd.in/d8maWHhM Full Paper: https://lnkd.in/dEENc5VD #LeanGPT #ActivityRecognition #KaizenCopilot