Welcome to Publishing Pulse, your weekly source for industry updates in online publishing. Stay informed about the latest trends and breakthroughs in the ad ecosystem, content creation, SEO, AI technology, and monetization.
If you prefer to listen to industry news, you can tune in to The Publisher Lab podcast. New episodes are released weekly on Thursday.
The Industry Is Unimpressed With Google Gemini
Google’s new generative AI model, Gemini, marks a significant step in the company’s AI development. Available in three variants, Gemini Ultra, Gemini Pro, and Gemini Nano, each version serves different purposes and audiences. Gemini Pro, primarily for English speakers in the U.S., is integrated into Google’s ChatGPT competitor, Bard. It offers enhanced reasoning and understanding capabilities and will become available to enterprise customers on December 13 through Vertex AI. Later, it’s expected to integrate into Google products like Duet AI, Chrome, Ads, and Search.
Gemini Nano, tailored for the upcoming Android 14 on Pixel 8 Pro, focuses on mobile-based applications like summarization in the Recorder app and generating suggested replies. Gemini models, trained on Google’s proprietary AI chips (TPUs), are touted for their efficiency and cost-effectiveness.
The most advanced, Gemini Ultra, is “natively multimodal,” meaning it can process and understand multiple data types, including text, audio, images, and videos. Despite its proficiency in various benchmarks, Google has been reticent about the specifics of its training data sources.
“I also think that Google has been focused on profits a little too much lately, you know, with their cash cow, Google search. And this is sort of counterintuitive to that. They basically have to cannibalize their search in order to get this [Gemini] to work. So now, they have to shift that profit mindset back into the best product for the customer or end user,” says Tyler Bishop on this week’s episode of The Publisher Lab podcast.
With Gemini’s introduction, concerns arise regarding its environmental impact, potential biases, and the handling of toxic content. The launch itself appears hasty, as pointed out by many publications and users across the ecosystem, leading to questions about Gemini’s full capabilities and the thoroughness of its development process. Google’s marketing strategy may have set lofty expectations for Gemini, potentially leading to initial disappointment in the product. The company is still exploring ways to monetize Gemini and fully harness its capabilities, with Gemini Ultra featuring a context window of approximately 24,000 words.
Google Confesses Altering Gemini Promotion Video
A promotional video for Google’s AI model, Gemini, has garnered significant attention on YouTube, amassing 1.6 million views. The video’s appeal lies in its portrayal of Gemini seemingly responding in real-time to spoken-word prompts and visual cues. However, Google later clarified that the video’s responses were accelerated for demonstration purposes and that the AI wasn’t actually responding to voice or live video input. Instead, the AI was prompted using still image frames from the video and accompanying text.
The video’s format, showing a person asking questions and presenting objects to the AI, created the illusion of real-time interaction. Notably, some of the video’s more remarkable moments—such as accurately identifying objects and engaging in a game—were facilitated by these still images and textual prompts, not live interactions.
Google explained that the demo aimed to test Gemini’s capabilities across various challenges by combining footage with text prompts. This revelation about Gemini’s capabilities, which are akin to those of OpenAI’s GPT-4, emerged amid a turbulent period in the AI industry. With Google reportedly working on advancing its AI technology further, it indicates a growing competition in the AI development space.
Google Project Ellmann: AI Tool Shows Users Their Life Story
Google’s “Project Ellmann” aims to use AI to create comprehensive narratives of users’ lives using mobile data. It plans to analyze photos and searches to tell personal stories. The project, utilizing Google’s advanced AI model Gemini, will process text, images, and audio to identify significant life events from users’ digital footprints.
A key feature, “Ellmann Chat,” resembles ChatGPT but with a focus on personal life knowledge, offering insights into users’ preferences and habits based on photo analysis. This initiative reflects a broader trend in tech, with companies like Google and Apple increasingly using AI to enhance personal memory creation and photo organization.
Despite the potential benefits, challenges such as accurate image recognition and balancing user privacy with AI capabilities remain central concerns for these evolving technologies.
AI Alliance Between Meta, IMB, Others
IBM and Meta have established the AI Alliance, a collective of over 50 organizations, including AMD, Intel, NASA, CERN, and Harvard University. This alliance is dedicated to fostering open innovation and science in the field of artificial intelligence, presenting a stark contrast to the more closed AI systems developed by entities like OpenAI and Google.
The AI Alliance emphasizes the importance of openness and transparency in AI development. It encourages the sharing of tools, knowledge, and research, particularly in the realms of AI model creation and training. Its membership is diverse, spanning across various sectors including technology, research, government, and academia.
Key initiatives of the alliance include setting development benchmarks, building an AI hardware accelerator ecosystem, supporting global AI research, and advocating for diversity in the development of AI foundation models. This formation of the AI Alliance marks a significant ideological divide in the AI community, between open and closed approaches to AI development. The long-term impact of this alliance on the trajectory of AI development remains to be seen.
OpenAI Responds To ChatGPT Getting ‘Lazier’
OpenAI has responded to user concerns about the perceived “laziness” of their GPT-4 model, clarifying that there have been no updates to the model since November 11th and that the issue was not intentional. They highlighted that no recent modifications have been made to ChatGPT or GPT-4 that would lead to this behavior.
The company acknowledged the unpredictability of the model’s behavior and is actively investigating the issue to find a solution. OpenAI emphasized that the model hasn’t undergone any autonomous changes since the mentioned date. However, they noted that variations in the model’s behavior can be subtle and may take time to both notice and rectify.
Meta Faces Potential New Data Use Limitations as FTC Gains Approval from Federal Judge
District Judge Timothy Kelly has chosen not to block the U.S. government from broadening a $5 billion privacy settlement with Meta, Instagram’s parent company. This decision empowers the Federal Trade Commission (FTC) to suggest new regulations, which could include prohibiting Meta from benefiting from data collected from users under the age of 18. These proposed rules could substantially affect Meta’s data-centric business model, especially as it focuses on younger demographics and delves into fields like virtual reality.
The FTC is also considering stricter regulations on facial recognition technology and halting new product releases until they pass third-party audits for privacy compliance. This move signifies one of the most considerable constraints on Meta’s operations since the 2020 privacy order resulting from the Cambridge Analytica scandal.
Meta intends to appeal this decision, disputing the FTC’s claims. Meanwhile, the FTC has set a deadline for Meta to respond to its proposed changes. The ruling grants the FTC the authority to reopen and modify the order concerning Meta without previous court approval, although Meta retains the option to contest any future FTC rule alterations in federal court.
