🚀 The AI Data Race has started

Also: Tech giants on a billion-dollar shopping spree

Author

Evolving AI
April 08, 2024

Welcome, AI enthusiasts

A new report reveals how leading AI companies are aggressively expanding their access to data, often overlooking copyright and licensing rights while collecting data to train their AI models. Moreover, these tech giants are on a billion-dollar shopping sprees, purchasing extensive collections of old images and videos to enhance AI learning. Also, Meta will soon implement "Made with AI" badges to label AI-generated content on all their platforms. Let's dive in!

In today’s insights:

  • The AI Data Race has started

  • The new Gold Mine called Data

  • Meta commits to transparency with a new labeling approach for AI-Generated content

Read time: 4 minutes

🗞️ LATEST DEVELOPMENTS

Source: Cath Virginia / The Verge

Evolving AI: Leading AI companies are disregarding copyright and licensing rights while gathering data to train their AI models

Key Points:

  • OpenAI's Whisper model transcribed over a million YouTube hours to train GPT-4.

  • Google and Meta also struggle with data limitations.

  • Legal and ethical questions abound amid aggressive data acquisition strategies.

Details:

As AI technologies keep advancing, their need for data keeps growing too. OpenAI, Google, and Meta partly ignored their guidelines and discussed intentionally violating copyrights, assuming their competitors would do the same. OpenAI used its Whisper model to transcribe a vast amount of content from YouTube to train its advanced model, GPT-4. Even though there were potential legal issues, OpenAI went ahead with this massive data collection by claiming it was fair use. At the same time, companies like Google and Meta are facing similar challenges. They are pushing the limits of what's acceptable for data use. Google subtly changed its privacy policies, while Meta thought about bold steps like buying a large publisher to legally access copyrighted content.

Our Thoughts:

Why is this aggressive data collection significant? As AI progresses, the ethical and legal standards lag behind. The key question is: how far can AI companies go before crossing a boundary, and what will the consequences be for privacy and copyright laws? This ongoing debate will shape the future of AI technology and how it is viewed by society. Companies are also testing synthetic AI-generated data for training, but this risks exacerbating existing errors and biases, potentially degrading performance over time. It also raises questions about legitimate data origins if models that generate artificial training data have been trained on copyrighted data.

Arbor is a AI summarization platform that brings the world's news to your fingertips, overcoming language and geographical barriers.

✅ Boost your reading efficiency by 10x

✅ Read the world in your own language

✅ 10,000+ topics summarized by AI

✅ Break videos into section summaries

✅ 100-word summaries for each topic

Source: Midjourney prompted by THE DECODER

Evolving AI: The old internet data is now a gold mine for AI.

Key Points:

  • Companies like Google and Meta spend a lot on old images and videos to help AI learn.

  • Prices change, with higher costs for special items like long videos or sensitive images.

  • There are legal issues as the FTC checks changes to user terms for selling data.

Details:

Photobucket, Shutterstock, and Freepik have turned their vast collections of old photos and videos into a profitable business by selling them to tech giants for AI training. This business involves various prices based on the type of content and who is buying it. Some items can sell for hundreds of dollars. Photobucket's CEO, Ted Leonard, is in the middle of complex talks with major tech companies to sell rights to billions of digital files. He's banking on a recent update to the terms of service that allows more freedom to make money from user content, a move that's being watched closely by regulators.

Our Thoughts:

Is data the new gold? As online archives become cash sources for AI training, the challenge is to balance innovation with users' rights. Will the push to use AI respect users' privacy and expectations, or are we moving towards a time when every upload adds money to corporate pockets?

Source: Meta | Labeling approach for AI-Generated content

Evolving AI: Meta updates its approach to AI-generated and manipulated media, enhancing transparency.

Key Points:

  • Introduction of "Made with AI" badges for deepfakes.

  • More contextual labels, fewer content removals.

  • Response to Oversight Board feedback.

Details:

Meta is updating how it handles AI-altered media by increasing the use of labels and cutting back on removing content. This change comes after feedback from its Oversight Board and new laws like the EU’s Digital Services Act. Starting in July, Meta will focus more on making things clear rather than just taking content down. It will add labels to give users better information about the AI content they see. This new way aims to balance the need for free speech with the fight against false information, especially important during elections.

Why It Matters:

As digital information gets more complicated, Meta’s move to add more details and context instead of just removing content is a careful step in handling digital truth and misinformation.

💡 Tip of the Day

But..what is a GPT really? This amazing new video provides a clear explanation of the technology behind Transformers, which are fundamental to large language models like ChatGPT. It breaks down how Transformers help computers understand and generate text by examining the relationships between words in a sentence, in terms that are easy to grasp.

🎯 SNAPSHOTS

Direct links to relevant AI articles.

🏥 ALS: AI-discovered drug for ALS enters clinical trials.

🇨🇦 Canada: Trudeau announces $2.4 billion for AI-related investments.

📈 Trending AI Tools

  • 🎤 CloneDub - A tool that converts audio files, YouTube links, and audio links into other languages (link)

  • 🕺 Rokoko - Create motion capture animations using your webcam (link)

  • 🛒 Wisely - A Google Chrome extension to help with Amazon shopping decisions through insightful product analysis (link)

  • 📊 Tome - Create amazing slide decks with AI (link)

  • 📈 Finalle.ai - Real-time data analysis of financial markets (link)

  • 🎨 Stylar.ai - AI-powered tool that transforms your photo into artwork in just a few clicks (link)

Reply

or to participate.