Image by Freepik
Meta Releases AI Model To Advance Virtual Agent Behavior In The Metaverse
- Written by Kiara Fabbri Former Tech News Writer
- Fact-Checked by Justyn Newman Former Lead Cybersecurity Editor
On Thursday, Meta FAIR has released several new research tools and findings aimed at advancing machine learning and artificial intelligence. These releases focus on areas such as agent development, robustness, safety, and machine learning architectures.
In a Rush? Here are the Quick Facts!
- Meta FAIR introduces research artifacts to enhance machine intelligence and improve AI development.
- Innovations include Meta Motivo for controlling virtual agents and Meta Video Seal for watermarking.
- Meta emphasizes democratizing access to advanced technologies for real-world interaction improvements.
Among the highlights are Meta Motivo, a foundation model for controlling virtual embodied agents, and Meta Video Seal, a video watermarking model designed to enhance content traceability.
Meta Video Seal builds on previous research in audio watermarking and enables the embedding of imperceptible watermarks in video content. The system is resistant to common modifications such as blurring, cropping, and compression, offering practical applications for safeguarding digital media.
Accompanying this is the Omni Seal Bench, a benchmarking platform for evaluating watermarking systems across different formats. This platform aims to foster collaboration within the research community.
Meta Motivo introduces a framework for unsupervised reinforcement learning. It uses a motion dataset to create a shared latent space for states, motions, and rewards.
The model demonstrates capabilities such as zero-shot motion tracking and goal-reaching while maintaining robustness against environmental variations like gravity and wind. These features have potential applications in virtual environments and animation.
Flow Matching, another release, provides an alternative to traditional diffusion methods for generative models. It supports various data types, including images, videos, and 3D structures, while improving computational efficiency and performance.
In the area of social reasoning, Meta Explore Theory-of-Mind presents a program-guided dataset creation method to train AI models for reasoning about beliefs and thoughts.
Initial tests indicate improvements in model performance on established benchmarks, with implications for enhancing reasoning in large language models.
Meta has also introduced Large Concept Models (LCMs), which aim to separate reasoning tasks from language representation by predicting conceptual ideas instead of individual tokens.
This approach reportedly improves tasks like summarization and multilingual processing. Additionally, the Dynamic Byte Latent Transformer eliminates the need for tokenization, offering more efficient processing of long sequences and rare text.
Other releases include Meta Memory Layers, which help scale the incorporation of factual knowledge into models, and tools for evaluating responsible image generation.
The integration of AI agents with physical-like bodies marks a significant shift in the metaverse, enabling more realistic interactions and dynamic virtual experiences.
However, these advancements could blur boundaries between the virtual and real worlds, raising questions about privacy, accountability, and the societal impact of increasingly lifelike virtual agents.
Image by Freepik
Australia To Impose Tax On Tech Giants Over Revenue-Sharing With News Outlets
- Written by Kiara Fabbri Former Tech News Writer
- Fact-Checked by Justyn Newman Former Lead Cybersecurity Editor
On Thursday, the Australian government announced plans to impose a tax on major digital platforms and search engines unless they agree to share revenue with Australian news media, as reported by the AP .
In a Rush? Here are the Quick Facts!
- Australia plans to tax tech companies unless they share revenue with media.
- The tax targets companies earning over 250 million Australian dollars annually from Australia.
- Meta, Google, and ByteDance are included in the new revenue-sharing initiative.
This is part of Australia’s ongoing efforts to regulate big tech companies. Last week, lawmakers passed a new law that prohibits children under 16 from using social media .
The new tax, set to take effect on January 1, will target tech companies generating over 250 million Australian dollars ($160 million) annually from Australian operations, according to Assistant Treasurer Stephen Jones and Communications Minister Michelle Rowland, as reported by the AP.
This includes companies like Meta, Alphabet (Google’s parent company), and ByteDance, the Chinese owner of TikTok.
The tax will be offset by payments to Australian media outlets, though the exact tax amount remains unclear. The government intends to make revenue-sharing the more cost-effective option.
“The news bargaining initiative will … will create a financial incentive for agreement-making between digital platforms and news media businesses in Australia,” Jones said as reported by Reuters .
The AP notes that a previous government introduced the News Media Bargaining Code in 2021, which required tech giants to negotiate revenue-sharing agreements with Australian media companies or face fines of 10% of their Australian revenue.
CBC pointed out that in August 2023, Meta blocked news links in Canada to avoid paying fees to media companies.
Since then, Canada has become the focal point of Facebook’s conflict with governments that have enacted or are considering laws requiring internet giants to compensate media companies for linking to news published on their platforms.
In response to the Australian ban Meta said in a statement that the current law was flawed. “The proposal fails to account for the realities of how our platforms work, specifically that most people don’t come to our platforms for news content” Meta statement said, as reported by CBC.
“News publishers voluntarily choose to post content on our platforms because they receive value from doing so,” Meta statement added,
CBC notes that Google has struck revenue-sharing agreements with more than 80 Australian news companies in the past three years and has committed to renewing those deals. But Google has raised doubts about the government’s new approach.
“The government’s introduction of a targeted tax risks the ongoing viability of commercial deals with news publishers in Australia,” a Google statement said, as reported by CBC.
TikTok noted that its users didn’t seek news. “As an entertainment platform, TikTok has never been the go-to place for news. We will actively engage in the consultation process and look forward to hearing more details,” a TikTok statement said, as reported by CBC.
Rowland said the revenue-sharing was needed to safeguard Australian democracy. “The rapid growth of digital platforms in recent years has disrupted Australia’s media landscape and it is threatening the viability of public interest journalism,” Rowland commented, as reported by CBC.
