OpenAI to solve AI bias

Reading time: 5 minutes  

 Newsletter | News | Tools 

Today we will discuss-

• 🤩OpenAI to create new, improved AI training datasets

• 🎬Hollywood actors secure safeguards against unauthorized AI

• 👨‍⚖️U.S. judge dismisses part of copyright lawsuit against Meta's Llama

• 😱13 amazing AI tools you might not have heard of

All this and more - Let's dive in!

👩‍🍳What’s cooking in the newsroom?

OpenAI seeks partnerships to create better training datasets for AI models

♨️News - On 9 November, OpenAI announced that it plans to collaborate with third-party organizations to create public and private datasets, for the training of AI models.

In a blog post, OpenAI explained that the initiative aims to involve more organizations in shaping the future of AI and benefiting from models that are 'more useful.'

🤔How does OpenAI plan to make the models more useful? It is no secret that AI models like ChatGPT are trained on open-source data from the Internet, which is flawed and contains biases. This is problematic because models tend to magnify these flaws and biases in their responses. Thus, to address this issue, OpenAI wants to create new and improved datasets that will ultimately make AI models more useful. 

The firm added that it also aims to broaden AI models' understanding across subjects, industries, cultures, and languages. To fulfill this, the company stated it will need extensive "large-scale" datasets that accurately represent human society and are currently not readily available online.

Furthermore, to ensure the quality of data, OpenAI said that it will remove sensitive and personal information, while also helping organizations digitize, structure, and remove errors from training data.

Hollywood actors’ strike ends with deal that protects against unauthorized use of AI images

☕News - After a 118-day strike, Hollywood actors have finally reached a tentative agreement with major studios that provides protection against the unauthorized use of AI-generated images. This addresses concerns about performers being replaced by "digital doubles."

Crabtree-Ireland, the lead negotiator for SAG-AFTRA, the union representing around 160,000 performers, mentioned that the newly established safeguards are designed to evolve alongside advancing technology, ensuring ongoing protection.

🕵️‍♂️What are these safeguards though? While we don't have all the details yet, what we do know is: 

• According to the agreement, movie studios are required to seek actors' permission before using their images in content generated by AI.

• Studios must also compensate performers whenever their digital doubles are featured on screen.

• Moreover, while the proposed agreement sets a minimum compensation level for AI uses, actors have the freedom to negotiate higher payments.

• Lastly, the proposed contract also includes safeguards around the use of generative AI to create synthetic actors.

Full details of the new contract will be disclosed after the SAG-AFTRA's national board votes on the proposal. Subsequently, the deal will proceed to union members for approval.

U.S. judge temporarily dismisses part of authors’ copyright lawsuit against Meta’s Llama

🏛️News - On 9 November, U.S District Judge Vince Chhabria announced his decision to temporarily dismiss a part of the copyright lawsuit filed by comedian Sarah Silverman and other authors against Meta's AI system Llama.

🧐Which parts of the lawsuit are being dismissed & why? In the lawsuit filed in July, authors claimed that the AI models' output violates their copyrights. Judge Chhabria expressed skepticism about this claim, questioning the notion that the text generated by Llama copies or resembles the authors' works. He even highlighted that queries to Llama don't seek copies or excerpts of specific books, citing comedian Sarah Silverman's book as an example.

In a separate claim, authors said that Llama itself is an infringing work. Judge Chhabria remarked that accepting this argument would mean Meta's Llama and Sarah Silverman's book are similar when placed side by side.

In line with this, the judge stated he would temporarily dismiss most of the claims, allowing the authors the opportunity to amend them. He further clarified that if the authors fail to demonstrate that Llama's output closely resembles their works, the dismissals would stand.

Here, it is important to note that despite seeking to dismiss most of the lawsuit previously, Meta hasn't challenged the authors' main claim that it violated their rights by using their books to train Llama. 

👩🏼‍💻What else is happening?

• City of Seattle in the U.S. releases generative AI policy

• OECD updates definition of Artificial Intelligence to inform EU's AI Act

• Nvidia plans to release three new AI chips for China

• Microsoft-backed Humane launches AI Pin, an OpenAI-powered wearable

• New AI noise-canceling headphone technology lets wearers pick which sounds they hear

📚AI comic books

• Cyberpunk: Peach John (2023) - A Japanese manga comic book that was illustrated using Midjourney, an online image generator. The story follows a young woman named Peach John who is trying to survive in a cyberpunk world.

• The Singularity (2019) - A comic book about the Singularity, a hypothetical moment in time when AI will surpass human intelligence. The story follows a group of scientists who are trying to prevent the Singularity from happening.

• Zarya of the Dawn (2022) - A short comic book about a non-binary person named Zarya who wakes up in an abandoned New York City. It was illustrated entirely using Midjourney, which later resulted in a copyright dispute.

👩🏼‍🚒Discover mind-blowing AI tools

1. Plai Day - An AI app that helps users enhance their videos with AI-generated effects and filters

2. BubbleTea - A comprehensive suite of apps for Slack that enhances team engagement and communication

3. Screenshot2Code - A tool that allows users to generate code from screenshots

4. UXAudit - A tool that provides instant feedback and suggestions to improve the usability, conversion, and messaging of a landing page

5. Feedeo - An interactive videos platform that allows users to create lifelike videos by uploading photos and filling out a script

6. Creative Fast AID - A tool that generates campaign ideas for NGOs and brands in minutes

7. BooksByAI - A platform that generates books written by AI based on users' interests

8. MealByMeal - A nutrition tracking tool that helps users log their meals, track their weight, and monitor their calorie and macro intake

9. ChatGitHub - A helpful information assistant designed to answer questions about GitHub

10. Omnidock - A tool that helps users find leads in private communities such as Facebook, Circle.so, and Slack

11. Excuse Generator - A fun tool that generates excuses for various situations

12. OnModel.ai - An app that allows fashion e-commerce stores to change models in their product photos using AI

13. ListingTurbo - An AI-powered tool that generates real estate listing descriptions

🐦AI Twitter influencers

• Demis Hassabis is the co-founder of DeepMind, which was acquired by Google in 2014. He is a leading AI researcher and is known for his work on AlphaGo, the first computer program to defeat a professional human Go player.

• Fei-Fei Li is a computer scientist and the director of Stanford's Human-Centered AI Institute. She is a pioneer in the field of computer vision and has made significant contributions to the development of deep learning algorithms.

• Max Tegmark is a physicist and cosmologist at MIT and the author of the book "Life 3.0: Being Human in the Age of Artificial Intelligence." Max is also a leading thinker on the future of artificial intelligence.