Developers build AI to read ancient scroll burnt in Mount Vesuvius eruption
Plus: US Space Force halts use of ChatGPT and more
AI in brief Computer scientists trained AI models to decipher words from parts of an ancient scroll preserved in thick layers of volcanic mud, where they laid for thousands of years after Mount Vesuvius erupted in 79 AD.
The charred script known as the Herculaneum scrolls was reportedly excavated in the 1700s, and is too fragile to handle and unroll. A team of researchers at the University of Kentucky's made 3D CT scans of the object to try and peer inside its contents, and managed to produce images of different fragments of the scroll at a four micrometer resolution.
Former GitHub CEO Nat Friedman launched a competition encouraging programmers to develop software that can help experts read the text. One contestant found a series of patterns that look like crackles in the script and led to the discovery of a first letter, prompting others to start finding more in an attempt to crack whole words.
Now, two students, Luke Farritor, and Youssef Nader, have won prizes for training AI models that revealed the ancient Greek word πορφυ̣ρ̣ας̣ meaning "purple," and others that look like ανυοντα meaning "achieving" and ομοιων meaning "similar". The latest attempt from Nader has uncovered four and a half columns of text that aren't completely legible yet.
The competition isn't over yet, and the top prize of $700,000 is still up for anyone who can devise a way to decipher four passages of text contained inside of two burnt-up but intact scrolls.
US Space Force temporarily bans ChatGPT
The US Space Force has urged staff to stop using generative AI tools like ChatGPT internally over data privacy fears. An internal memo that was reportedly sent to all military personnel, collectively referred to as Guardians, said that they were temporarily banned from using applications like ChatGPT until the Space Force's Chief Technology and Innovation Office approves the technology.
The memo led by Lisa Costa, the Space Force's chief technology and innovation officer acknowledged that LLMs will help make Guardians more productive, but raised "data aggregation risks". Models can be trained on data generated in conversations, and it's possible that if the military divulges secret data users could extract or uncover the information in the future.
"A strategic pause on the use of generative AI and Large Language Models within the US Space Force has been implemented as we determine the best path forward to integrate these capabilities into Guardians' roles and the USSF mission," Air Force spokesperson Tanya Downsworth told Reuters in a statement. "This is a temporary measure to protect the data of our service and Guardians," she added.
The US Space Force has joined a generative AI task force with other US Department of Defense agencies to probe its security risks and develop best practices to deploy the technology responsibly.
- DeepMind AI tool helps historians restore ancient texts
- Apple becomes the latest company to ban ChatGPT for internal use
- Italy will say ciao to ChatGPT ban if OpenAI does indeed think of the children
- How 'AI watermarking' system pushed by Microsoft and Adobe will and won't work
Google is using AI and Maps to make traffic lights more efficient
Google has been working with city planners across the world to better control traffic lights to help drivers reach their destinations more efficiently and reduce carbon emissions.
Researchers working on the company's Project Green Light continue to analyze the flow of traffic data using Google Maps with AI. They have built an AI-based model of intersections to recommend how to synchronize traffic lights to help more cars pass through more smoothly with fewer stops and starts.
The models look at driving patterns, light scheduling, and how they affect one another. These recommendations are then passed to officials, who can alter the timing of different sets of traffic lights. Project Green Light is deployed across 70 intersections in 12 cities from Haifa, Israel, to Rio de Janeiro, Brazil, to Bangalore, India.
"We're excited to share that early numbers indicate a potential for up to 30 percent reduction in stops and up to 10 [percent] reduction in emissions at intersections…In the intersections where Green Light is already live, this can save fuel and lower emissions for up to 30 [million] car rides monthly," Google said in a blog post.
The company has invited more cities to sign up and join the initiative. ®