UPP News: What's new in the AI field?

Subscribe to our newsletter and stay up-to-date with the latest AI news updates.

1. Google’s new AI will play video games with you - but not to win.

Source: The Verge

Google DeepMind has introduced SIMA, an AI agent designed to learn gaming skills to play more like a human rather than an overpowered AI. SIMA, or Scalable, Instructable, Multiworld Agent, is still in the research phase and aims to eventually master any video game, including open-world and non-linear games. It combines natural language instruction with 3D world understanding and image recognition.

Unlike traditional game AI, SIMA is not trained to win but rather to follow instructions and play alongside human players. Google collaborated with eight game developers to train and test SIMA, using games like No Man’s Sky, Teardown, Valheim, and Goat Simulator 3. To teach SIMA, Google recorded human players giving instructions while others played independently, capturing language instructions and resulting actions. This data was used to train SIMA to predict gameplay outcomes. Currently, SIMA has about 600 basic skills, but it aims to perform more complex tasks in the future. While not intended to replace existing game AI, SIMA is envisioned as another player that complements human players.

It's too early to determine its broader applications in gaming, but it may eventually evolve to perform more complex tasks and communicate with players.

2. China uses AI to manage the world's largest high-speed railway system.

Source: VTC news

China has revolutionized railway management by implementing advanced artificial intelligence (AI) technology across its expansive high-speed railway network, the largest in the world. This AI system excels at accurately predicting errors and issuing timely warnings to prevent potential incidents, ensuring the smooth operation and safety of the railway lines. Engineers involved in the project have reported an impressive 80% reduction in minor rail faults within just one year of implementing the AI system, highlighting its effectiveness in enhancing maintenance and operational efficiency.

Operating at an astounding length of 45,000 kilometres, China's high-speed railway network is centrally controlled by the AI system headquartered in Beijing. The system continuously processes vast amounts of real-time data, enabling it to swiftly detect and respond to anomalies within 40 minutes. This proactive approach to maintenance has garnered praise from industry experts for its transformative impact on railway safety and operational reliability.

While China leads the way in adopting AI for railway management on a large scale, other countries face challenges with inadequate maintenance practices, resulting in safety risks and operational disruptions. However, nations like Germany and Switzerland are also exploring AI-driven solutions to enhance the efficiency of their railway networks, albeit on a smaller scale compared to China's ambitious endeavors.

3. The AI video creation tool Sora will soon be widely released.

Source: The Verge

OpenAI is on the verge of launching Sora, a text-to-video tool, for commercial use, with a pricing structure akin to its existing product, Dall-E. Despite its initial availability being limited to select users for testing purposes, Sora has managed to capture significant attention within the creative community for its unique ability to transform textual inputs into compelling videos. Building on this momentum, OpenAI is poised to further refine Sora's capabilities by integrating sound-generation features, aiming to elevate the realism of the generated content.

While OpenAI has been somewhat reserved regarding the specifics of the data used to train Sora, the company's Chief Technology Officer, Mira Murati, has affirmed that the training data primarily consists of licensed or publicly available datasets. In parallel, OpenAI is proactively addressing potential ethical implications associated with AI-generated content. The company plans to implement measures such as labelling AI-generated videos and collaborating with experts to assess the risk of misinformation and bias.

Looking ahead, OpenAI is committed to ensuring responsible usage of Sora. This involves imposing restrictions on the creation of images featuring celebrities, aligning with the policies applied to Dall-E. Additionally, visual cues will be integrated into AI-generated videos to differentiate them from authentic content. Through these initiatives, OpenAI aims to uphold ethical standards while fostering public trust in AI technologies.

4. Alibabas AI transforms static images into ones that can speak and sing.

Source: Times of India

Alibaba's Emotive Portrait Alive (EMO) AI, developed by the Institute of Intelligent Computing, represents a significant advancement in the realm of AI-generated content. Unlike previous models, EMO possesses the remarkable ability to imbue static images with lifelike qualities, enabling them to speak and sing realistically. This breakthrough is made possible by EMO's capacity to generate natural facial expressions, synchronize lip movements with accompanying sound, and create dynamic videos that capture the essence of the input images.

What sets EMO apart from its predecessors is its synthetic approach, which avoids the need for intermediate 3D models or facial landmarks. Instead, the AI seamlessly transitions between frames, preserving consistency in the video and delivering content that is highly expressive and true to life. This methodological innovation, combined with extensive training on large datasets of images, sounds, and videos, empowers EMO to produce dynamic and engaging content across various languages.

Alibaba's commitment to transparency is evident in the release of EMO's data on Github and the publication of research papers on ArXiv. However, details regarding the general release of this groundbreaking AI remain undisclosed. Nevertheless, EMO's emergence marks a significant leap forward in AI technology, promising exciting possibilities for creating compelling visual content from static images.

?Interested in AI Implementation for your business? Let us reach you here!

?