•
February 8, 2024
•
7 mins
Hello Niuralogists!
In the dynamic field of artificial intelligence, this week's edition is dedicated to presenting the latest breakthroughs. Our primary goal is to examine the influence of these advancements on various aspects of our lives, spanning workplaces, businesses, policies, and personal encounters. This edition will delve into recent developments like AI uncovering ancient texts in the Vesuvius Challenge the exposed rebranding of Bard and the launch of Gemini Ultra in a leaked document.
For a more in-depth understanding, keep on reading...
AI has played a pivotal role in unraveling the contents of a charred, 2,000-year-old scroll from Herculaneum's expansive Roman library, preserved beneath the ashes of Mount Vesuvius' eruption in 79 A.D. Former GitHub CEO Nat Friedman sponsored a $1 million contest challenging teams to extract Greek text from scans of the damaged scrolls, known as the 'Herculaneum papyri.' Despite being excavated in the 18th century, the scrolls remained indecipherable due to the eruption-induced damage. The software leveraged AI algorithms to identify patterns overlooked by human eyes, resulting in the revelation of over 2,000 readable letters. This achievement underscores the profound impact of AI not only on shaping the future but also on unveiling and reshaping our understanding of the past, offering a newfound ability to decipher texts and artifacts previously lost to the annals of time, thereby potentially rewriting the narrative of history.
A leaked changelog circulating on social media has unveiled Google's strategic rebranding of its Bard chatbot to Gemini, coinciding with the highly anticipated launch of the Gemini Ultra model. The leaked document indicates that the transition from Bard to Gemini will occur very soon, aligning the brand with its underlying generative models. Additionally, a paid 'Gemini Advanced' tier, leveraging the more powerful Ultra model, is set to launch on the same date. Gemini's expansion includes the introduction of its first Android app on select Pixel and Galaxy devices, with global expansion planned. Notably, Canada gains access to Gemini on the web, with App access soon to follow. While the rebranding aligns with previous speculations, the excitement surrounds the Ultra's launch, as its capabilities could potentially rekindle the LLM acceleration race and prompt OpenAI to expedite the rollout of GPT-5.
The UK armed forces and Defense Science and Technology Laboratory (Dstl) recently collaborated with the military forces of Australia and the US as part of the AUKUS partnership in a groundbreaking trial focused on advancing AI and autonomous systems. Named Trusted Operation of Robotic Vehicles in Contested Environments (TORVICE), the trial took place in Australia under the AUKUS partnership formed last year. It aimed to assess the resilience of robotic vehicles and sensors in scenarios involving electronic attacks, GPS disruption, and other threats, providing valuable insights into the reliability of autonomous systems crucial for future military operations. The trial featured US and British autonomous vehicles conducting reconnaissance missions while Australian units simulated battlefield electronic attacks. The analysis of performance data will contribute to enhancing protections against system failures or disruptions, marking a significant step in harnessing AI and autonomy to reduce risks for soldiers and revolutionize military capabilities. This collaborative effort among AUKUS allies accelerates development in this critical area of warfare, fostering interoperability and strengthening deterrence in the Indo-Pacific region.
Microsoft has unveiled a series of collaborations with media organizations aimed at integrating AI tools and offering guidance on utilizing generative models for news gathering and operational processes. As part of these partnerships, Microsoft will sponsor Semafor's 'Signals' feed, employing AI to research and curate insights while maintaining journalist-authored content. Semafor's writers will also harness AI to sift through information from global news sources and diverse languages, enhancing research capabilities. Additional partnerships involve CUNY’s journalism school, The Online News Association, The GroundTruth Project, and Nota. This announcement follows the New York Times' lawsuit against OpenAI, alleging the training of ChatGPT on its news articles. The collaboration reflects a strategic move by media organizations to leverage AI for tasks beyond writing, acknowledging the occasional challenges posed by AI-generated text and aiming to streamline various aspects of their work.
Apple has introduced a groundbreaking open-source AI model named "MGIE," short for MLLM-Guided Image Editing, designed for instruction-based image editing. Developed in collaboration with researchers from the University of California, Santa Barbara, MGIE employs multimodal large language models (MLLMs) to interpret natural language commands and execute pixel-level manipulations. The model, presented at the International Conference on Learning Representations (ICLR) 2024, demonstrates improved automatic metrics and human evaluation, maintaining competitive inference efficiency. MGIE integrates MLLMs into the image editing process, deriving expressive instructions from user input and generating visual imagination for desired edits. This versatile model handles a spectrum of editing scenarios, including Photoshop-style modifications, global photo optimization, and local edits. Available as an open-source project on GitHub, MGIE facilitates easy use and customization for various image editing tasks, showcasing Apple's prowess in AI research and development. While representing a significant breakthrough, experts note ongoing efforts to enhance multimodal AI systems, signaling the rapid evolution of assistive AI in creative endeavors.
📬 Receive our amazing posts straight to your inbox. Get the latest news, company insights, and Niural updates.
MIT researchers have explored how incorporating symmetries within datasets can significantly reduce the amount of data required for training neural networks. Led by MIT PhD student Behrooz Tahmasebi and advised by Stefanie Jegelka, an associate professor in EECS and affiliate of CSAIL and the MIT Institute for Data, Systems, and Society, the research involved modifying Weyl's law to factor in symmetry when assessing dataset complexity. This innovative approach, presented at the Neural Information Processing Systems conference, offers insights into improving sample complexity and achieving exponential gains by exploiting symmetries, potentially revolutionizing machine learning tasks that involve intricate datasets.
The latest enhancements in Google Bard, featuring an exciting addition that allows users to generate images through casual conversation within a chat, are to be discovered. Simply follow these straightforward steps to unlock your creative possibilities: Head to Bard, keeping in mind that Bard's image generation is accessible in most countries, excluding the EU and Canada. Engage in a chat with Bard, instructing it to "Generate an image of [describe your image]." Feel free to ask Bard for a prompt to kickstart the process. Bard will promptly provide two images based on your instructions. Download the images, explore more possibilities, or refine your prompt to further enhance the generated images.
Started by Sync Labs, there's a cool way to change videos using smart computer tools. These tools, like turning words into text, changing text into talking and making sure the talking matches the mouth movements, work together to make videos in different languages. Here's how it goes: sign up for Sync Labs for free. Then, put in a video or a YouTube link. It's best if the video has just one person talking and is not super high quality. Pick the language you want, and click submit. In a few minutes, you'll see the main person in your video talking in the new language, and it looks just right.
📸 Kondor for iOS is ChatGPT for your phone camera
🎥 Pictory GPT effortlessly creates captivating videos
📊 Matrices simplify data analysis and automate tasks for better insights
📚 Books GPT is your personal AI guide for discovering and recommending books
🛠️ Promptly builds generative AI applications without coding experience
