The Week in Generative AI: August 16, 2024 edition

Welcome to The Week in Generative AI, a weekly roundup for marketers from Quad Insights that sums up the latest news surrounding this rapidly evolving technology. 

Google just announced that Gemini AI features are now available on Android devices, aiming “to put its AI in front of consumers before Apple, which will launch its AI on iPhones, Macs and iPads later this year,” CNBC’s Jennifer Elias reports. With this Android software update, Gemini AI will allow users the ability to ask the assistant questions, check calendar availability across apps and more, Elias notes, adding that the tech giant also announced the launch of the Google Pixel 9 series of smartphones, which come equipped with Gemini AI capabilities including AI photo editing.

Previously: “AI or DIY? Google Gemini ad controversy raises important questions for content creation,” from last week’s edition of this column.

See also: “Apple releases first preview of Apple Intelligence,” from the Aug. 2 edition of this column.

In a suit brought by artists alleging that Stability AI used billions of copyrighted works without permission to train its text-to-image AI model Stable Diffusion, U.S. District Court Judge William Orrick on Monday advanced all the copyright infringement and trademark claims, propelling the case to discovery phase, The Hollywood Reporter’s Winston Cho reports. The decision to move forward with discovery requires Stability AI to provide further details related to how copyrighted materials are collected for large language model (LLM) training, Cho notes, adding that this could also implicate other companies whose AI products were built on the Stable Diffusion model.

Previously: “Report: Apple, Nvidia and others used YouTube subtitles to train AI without permission,” from the July 19 edition of this column.

More AI controversy:

• “MIT researchers release a repository of AI risks” (TechCrunch)
• “Ex-Google CEO says successful AI startups can steal IP and hire lawyers to ‘clean up the mess’” (The Verge)

In a new report, OpenAI has outlined risks associated with its newest iteration of ChatGPT, including “oddities in ChatGPT-4o technology such as the ability to clone a voice and finish thoughts and sentences,” MediaPost’s Laurie Sullivan reports. Risks discovered by OpenAI in testing include unauthorized cloning of voices (including the user’s), inadvertently finishing sentences and outputting of “nonverbal vocalizations,” such as moans, screams and gunshots, Sullivan notes, adding that the report also lays out controls OpenAI has instituted to mitigate these risks.

Previously: “Scarlett Johansson threatens legal action against OpenAI for allegedly copying her voice,” from the May 24 edition of this column.

Previously: “Did ChatGPT just lose its mind?” from the Feb. 23 edition of this column.

See also: “When AI goes mad: Models suffer without fresh data, study finds,” from last week’s edition of this column.

“The most important takeaway from our work is that we cannot yet fully trust the outputs of model generations. At present, even the best models can generate hallucination-free text only about 35% of the time.”

—Cornell University doctoral student Wenting Zhao in an interview with TechCrunch about recent academic research that fact-checked outputs from 15 large language models.

xAI’s newest iteration of chatbot Grok can now generate AI images on X, and as some users have used the tool to create sensitive material, the image generator’s guardrails — or seeming lack thereof — have come into question, The Verge’s Adi Robertson reports. While Grok’s stated content guidelines are similar to those of other companies, it has generated numerous images that other AI tools, such as OpenAI’s ChatGPT, would not, Robertson notes. In testing this theory, The Verge found that Grok accepted prompts leading to images of Taylor Swift in lingerie, Bill Gates doing cocaine, and more.

Previously: “xAI open sources Grok” (The Verge)

Previously: xAI raised $6 billion in a recent funding round, as we noted in the May 29 edition of The Weekly 10.

A new National Research Group survey commissioned by Google Cloud attempts to shed light on AI’s current ROI opportunity, Consumer Goods Technology’s Lisa Johnston reports. Per the announcement of the survey’s findings, 86% of AI early-adopter companies across industries reported revenue increases estimated at more than 6%. Additionally, 61% of respondents said they have at least one generative AI application in production, Johnston notes. Find the full story here.

Previously: “Goldman Sachs sees challenges ahead for AI ROI,” from the July 19 edition of this column.

More AI trends:

• “McKinsey Sees AI Surge Boosting Consulting Amid Industry Slump” (Bloomberg)
• “How AI Is Transforming the U.S. Retail Delivery and Supply Chain Landscape” (Consumer Goods Technology)

AI tools and advancements:

• “Google brings AI answers in Search to new countries” (Reuters)
• “YouTube Expands AI Topics in Live Streams” (Social Media Today)

AI strategy:

• “Nvidia Looks To Disrupt The Health Care Industry With AI” (Investor’s Business Daily)