OpenAI’s Sam Altman Revealed as One of Reddit’s Biggest Shareholders

Reddit's journey towards a stock market listing significantly highlights its strategic relationships with AI vendors, including potential agreements with companies like OpenAI. In its IPO prospectus, Reddit disclosed entering into data licensing agreements valued at $203 million, anticipating at least $66.4 million in revenue for 2024. These partnerships, possibly including one with an unnamed large AI company rumored to be Google, underscore the value of Reddit's vast trove of data—over 1 billion posts and 16 billion comments—for training AI models. Reddit's pivot to monetizing its data comes as AI vendors seek legally sound means to access diverse datasets, moving away from unrestricted scraping practices.

Reddit's data, encompassing a wide array of real-time discussions on topics ranging from sports to the latest trends, is crucial for training large language models (LLMs) that power applications such as chatbots and content generation tools. This shift towards data licensing reflects a broader trend where content creators seek compensation from AI companies that use their content to train models. These developments are part of a larger discourse on the ethical and legal dimensions of data use in AI, with companies like OpenAI securing agreements with various content providers to ensure their models are trained on legally sourced data.

The IPO filing also revealed interesting insights into Reddit's financials and ownership structure, with OpenAI CEO Sam Altman holding an 8.7% stake, making him the third-largest shareholder. Reddit's emphasis on AI in its prospectus, including a significant revenue projection from data licensing, indicates the company's strategic focus on leveraging its unique dataset to fuel AI advancements. This move not only capitalizes on the growing demand for high-quality training data but also aligns with broader industry trends toward ethical data use and the creation of more sophisticated AI models.

Google pauses AI tool Gemini’s ability to generate images of people after historical inaccuracies

Google has temporarily halted the functionality of Gemini, its advanced generative AI model, from creating images of people. This decision comes in response to the tool generating historically inaccurate images, such as the U.S. Founding Fathers depicted as individuals of different racial backgrounds, which has sparked criticism and ridicule on social media. The company aims to update the technology to enhance the accuracy of human depictions and plans to re-release an improved version soon.

The controversy surrounding Gemini's output highlights the challenges faced by generative AI technologies in producing culturally and historically sensitive content. Despite the tool's capacity to generate a diverse array of human images, which is generally viewed positively for its inclusivity, its misrepresentation of historical figures has led to backlash. Google has acknowledged these issues and is committed to addressing the inaccuracies to ensure more respectful and accurate depictions.

This incident reflects broader concerns within the AI community about bias and the ethical implications of AI-generated content. Previous instances, such as a 2015 controversy where Google's AI misclassified Black men, underscore the importance of developing AI technologies that are both accurate and sensitive to diversity, equity, and inclusion. As AI tools become more integrated into our daily lives, the need for vigilant oversight and continuous improvement to prevent such errors becomes increasingly critical.

Stable Diffusion 3 arrives to solidify early lead in AI imagery against Sora and Gemini

Stability AI has unveiled Stable Diffusion 3 (SD3), their latest image-generating AI model, designed to compete with new offerings from OpenAI and Google. SD3 introduces an updated architecture, including an advanced "diffusion transformer" and a new technique called "flow matching," aimed at improving quality without significant overhead. The model is versatile, designed to run on a range of hardware setups, and spans from 800 million to 8 billion parameters. Notably, SD3 emphasizes multimodal understanding and potential for video input and generation, setting it apart from its API-limited competitors.

The emphasis on upgrading tooling and lowering barriers to use signifies Stability AI's ambition to become an indispensable white-label generative AI platform. In contrast to its competitors, Stable Diffusion aims for widespread applicability and minimal restrictions on content generation. Importantly, the announcement highlighted a commitment to safety, detailing proactive measures to prevent misuse by bad actors through continuous collaboration with experts and the community. These safeguards are a critical component of the model's development and deployment process.

As Stable Diffusion 3 gears up for its public release, it's positioned as a significant player in the AI-generated imagery field, promising enhanced capabilities and broader accessibility. While details on the technical advancements and specific safeguards remain sparse, the anticipation builds for a more inclusive and safe environment for AI-driven content creation. Stability AI's focus on safety, combined with its ambitious technological strides, underscores its challenge to competitors and its potential to redefine the landscape of generative AI.

Chrome gets a built-in AI writing tool powered by Gemini

Google Chrome has introduced a novel AI writing generator, leveraging its Gemini AI model to expand the "Help me write" feature from Gmail to the entire web. This tool, still in the experimental stage, allows users to generate or rewrite text across any web page in English on Windows, Mac, and Linux platforms. By enabling the feature through Chrome's "Experimental AI" settings, users can access it by right-clicking any text field. Despite its capabilities, the tool is designed primarily for short-form content, without offering enhanced functionalities to Gemini Advanced subscribers.

The AI generator is context-aware, tailoring suggestions based on the website's content to provide more relevant recommendations. Google's engineering director, Adriana Porter Felt, highlighted this feature, noting its ability to enhance content creation, such as pulling key product features to support reviews. Users can also adjust the output's length and tone, enhancing the tool's versatility for various writing tasks.

Google has emphasized privacy considerations, stating that text, content, and URL data used by the service will be processed under its existing privacy policy to improve the feature. This includes contributions to generative model research and machine learning technologies, with a human review process in place to oversee the development and application of this AI tool, underscoring Google's commitment to responsible AI use and user privacy.

Other stuff

• Windows is getting its own Magic Eraser to AI-modify your photos

• An ‘education legend’ has created an AI that will change your mind about AI

• Tyler Perry Puts $800M Studio Expansion on Hold After Seeing OpenAI’s Sora: “Jobs Are Going to Be Lost”

• New report: 60% of OpenAI model's responses contain plagiarism

• How many news websites block AI crawlers?

• AI’s Unlikely Benefactor: Blackstone’s 77-Year-Old CEO Steve Schwarzman

• Gen-Z is now using deep fakes to teach each other calculus 

• A friendly reminder that anything built on OpenAI's GPT4 via API is vulnerable to invisible prompt injection via Unicode tags.

• Some mind-blowing examples of things you can do with Google Gemini 1.5 Pro.

  • Collect and Organize tweets in JSON formats from a video

  • Generate Selenium code to replicate a task recorded by video

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.