- The Rundown AI
- Posts
- AI deciphers 2,000-year-old ancient scroll
AI deciphers 2,000-year-old ancient scroll
PLUS: Meta takes on deepfakes
Rowan Cheung
February 07, 2024
Welcome, AI enthusiasts.
AI may be the future — but it’s also helping us unlock the past.
A charred, 2,000-year-old ancient scroll was just deciphered using new AI algorithms, a triumph that could rewrite history as we know it. Let’s get into it.
In today’s AI rundown:
AI helps reveal ancient Herculean texts
Meta takes on AI image labeling
How to generate HTML from screenshots using GPTs
Research enables more agile, safe robot movement
8 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
AI & HISTORY
Image source: Vesuivius Challenge
The Rundown: AI algorithms just unlocked text from a charred, 2,000-year-old scroll from Herculaneum's vast Roman library, buried by the eruption of Mount Vesuvius' in 79 A.D.
The details:
A $1M contest sponsored by former GitHub CEO Nat Friedman challenged teams to extract Greek text from scans of damaged scrolls called the ‘Herculaneum papyri’.
Hundreds of the scrolls were excavated in the 18th century but were previously undecipherable due to damage from the eruption.
The winning software used AI to detect patterns missed by humans, revealing over 2,000 readable letters.
Why it matters: While AI is being hailed as the future — don’t underestimate its ability to uncover the past. The newfound ability to decipher texts and artifacts previously lost to time could completely rewrite our knowledge of history.
TOGETHER WITH CHIMA
The Rundown: One place to get immediate intelligent support and recommendations for any product or service.
Instant Solutions: Get immediate personalized support for any product or service issue.
Expert Access: Access specialized support in engineering, legal, tax, healthcare, etc.
Product Wisdom: Track support metrics, peer reviews, and insights for companies and their offering.
META
Image source: Meta
The Rundown: Meta just published a blog outlining new efforts to label AI-created images and media across its platforms while pushing for common standards around identifying synthetic content.
The details:
Meta plans to add labels to photorealistic AI images posted on Facebook, Instagram, and Threads over the next year using invisible watermarks and metadata.
The company co-founded technical standards for AI markers and is currently building tools to detect the metadata at scale.
Meta is adding user disclosure requirements for realistic images, videos, and audio posts that lack identifiers, with potential penalties for violations.
Why it matters: While AI deepfakes and deceptive media are clearly a massive problem — it feels like an issue that is going to be near impossible to regulate. Meta’s solutions are a step in the right direction, but workarounds will likely pop up that turn enforcement into a game of whack-a-mole.
AI TRAINING
The Rundown: With ChatGPT’s vision capabilities and custom GPTs, you can now generate code for a website by simply providing a screenshot.
Step-by-step:
Head over to the GPT store and search for the Screenshot To Code GPT.
Upload a screenshot of a website or application.
The GPT will generate the HTML code using Tailwind CSS to replicate the layout and style of the webpage in the screenshot you have provided.
You can now use this code to create your version of the website and even get ChatGPT to help you with the rest of the setup or any potential changes/configurations — just ask!
AI RESEARCH
Image source: Carnegie Melon University and ETH Zurich
The Rundown: Researchers at Carnegie Melon University and ETH Zurich just developed a framework allowing four-legged robots to navigate cluttered spaces at high speeds of nearly 7 MPH without collisions.
The details:
An agile program enables the robot to move rapidly around obstacles, while another program stops it if dangerous situations arise.
A special prediction function decides when to switch between fast and safe, helping better guide the robot’s motion.
A perception system predicts where obstacles are from camera angles to allow the robot to react in real time.
The complete system was tested on a Unitree Go1 robot moving through tight and crowded real-world spaces.
Why it matters: As robots are deployed into more complex situations, the ability to perform quick, agile, and safe movements will be absolutely necessary. This research could help unlock new models for building more dynamic robots to better navigate the chaos of the real-world.
NEW TOOLS & JOBS
🗣️ Deepgram- Introducing Aura, a powerful text-to-speech (TTS) API built for real-time conversations. Be the first to try Aura (link)*
🤖 CoachVox- Create an AI version of yourself to engage with your audience (link)
📝 GrammarBot- Write flawless English with AI-powered grammar and spelling correction (link)
👨🏻💻 code GPT- Assists with coding by providing real-time suggestions and guidance (link)
📅 BeforeSunset AI- Focus on completing your tasks effortlessly with automated daily planning (link)
📚 Physics GPT- Expert insights in advanced physics (link)
📄 Theodore AI- Transform complex text into simple content with real-world examples (link)
🔭 Galileo AI- Generate UI designs at lightning speed with AI, now open to the public (link)
PS — we're hiring at The Rundown.
*Sponsored listing
QUICK HITS
Stanford University and Stability AI introduced CheXagent, a new model for automated Chest X-Ray interpretation that outperforms existing models in clinical tasks.
Google’s Lookout App released a new feature called ‘Image Question and Answer’, using AI to provide audio responses describing an uploaded image to help blind and partially-sighted people better perceive the world.
OpenAI also released new AI image labeling efforts, with generations from ChatGPT and GPT APIs now including C2PA metadata — though the company notes the metadata can still be removed.
Tencent released a new version of its DynamiCrafter video generation model, showcasing enhanced capabilities and upgraded resolution.
Britain announced a $125M investment to educate regulators on AI and establish nine new AI research hubs, focusing on areas including healthcare, chemistry, mathematics, and responsible AI.
New research found that a ChatGPT-like system with minimal adjustments showed impressive capabilities in predicting chemical properties and reactions.
Life Corporation, a Texas firm already caught in past illegal call schemes, was found to be behind recent AI-generated robocalls of President Biden telling New Hampshire primary voters to skip the election.
THAT’S A WRAP
SPONSOR US
Get your product in front of over 450k+ AI enthusiasts
Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.
FEEDBACK
How would you rate today's newsletter?Vote below to help us improve the newsletter for you. |
If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.
roll
Join the conversation