\nGoogle brings new generative models to Vertex AI, including Imagen<\/br>
\n2023-05-10 22:13:12<\/br><\/p>\n
To paraphrase<\/a> Andreessen Horowitz, generative AI, particularly on the text-to-art side, is eating the world. At least, investors believe so \u2014 judging by the billions of dollars they\u2019ve poured into startups developing AI that creates text and images from prompts.<\/p>\n Not to be left behind, Big Tech is investing in its own generative AI art solutions, whether through partnerships with the aforementioned startups or in-house R&D. (See: Microsoft teaming up with OpenAI for Image Creator.<\/a>) Google, leveraging its robust R&D wing, has decided to go the latter route, commercializing its work in generative AI to compete with the platforms already out there.<\/p>\n Today at its annual I\/O developer conference, Google announced new AI models heading to Vertex AI<\/a>, its fully managed AI service, including a text-to-image model called Imagen. Imagen, which Google previewed via its AI Test Kitchen<\/a> app last November, can generate and edit images as well as write captions for existing images.<\/p>\n \u201cAny developer can use this technology using Google Cloud,\u201d Nenshad Bardoliwalla, director of Vertex AI at Google Cloud, told TechCrunch in a phone interview. \u201cYou don\u2019t need to be a data scientist or developer.\u201d<\/span><\/p>\n Getting started with Imagen in Vertex is, indeed, a relatively straightforward process. A UI for the model is accessible from what Google calls the Model Garden, a selection of Google-developed models alongside curated open source models. Within the UI, similar to generative art platforms such as MidJourney and Nightcafe, customers can enter prompts (e.g. \u201ca purple handbag\u201d) to have Imagen generate a handful of candidate images.<\/p>\n Editing tools and follow-up prompts refine the Imagen-generated images, for example adjusting the color of the objects depicted in them. Vertex also offers upscaling for sharpening images, in addition to fine-tuning that allows customers to steer Imagen toward certain styles and preferences.<\/p>\n As alluded to earlier, Imagen can also generate captions for images, optionally translating those captions leveraging Google Translate. To comply with privacy regulations like GDPR, generated images that aren\u2019t saved are deleted within 24 hours, Bardoliwalla says.<\/span><\/p>\n \u201cWe make it very easy for people to start working with generative AI and their images,\u201d he added.<\/span><\/p>\n Of course, there\u2019s a host of ethical and legal challenges associated with all forms of generative AI \u2014 no matter how polished the UI. AI models like Imagen \u201clearn\u201d to generate images from text prompts by \u201ctraining\u201d on existing images, which often come from data sets that were scraped together by trawling public image hosting websites. Some experts suggest that training models using public images, even copyrighted ones, will be covered by the fair use doctrine<\/a>\u00a0in the U.S. But it\u2019s a matter that\u2019s\u00a0unlikely<\/a> to be settled anytime soon.<\/p>\n Google\u2019s Imagen model in action, in Vertex AI.<\/p>\n<\/div>\n To wit, two companies behind popular AI art tools, Midjourney<\/a>\u00a0and\u00a0Stability AI<\/a>, are in the crosshairs of a\u00a0legal case<\/a>\u00a0that alleges they infringed on the rights of millions of artists by training their tools on web-scraped images.\u00a0Stock image supplier Getty Images has taken Stability AI to court, separately, for\u00a0reportedly<\/a>\u00a0using millions of images from its site without permission to train the art-generating model\u00a0Stable Diffusion.<\/p>\n I asked Bardoliwalla whether Vertex customers should be concerned that Imagen might\u2019ve been trained on copyrighted materials. Understandably, they might be deterred from using it if that were the case.\u00a0<\/span><\/p>\n Bardoliwalla didn\u2019t say outright that Imagen wasn\u2019t trained on trademarked images \u2014 only that Google conducts broad \u201cdata governance reviews\u201d to \u201clook at the source data\u201d inside its models to ensure that they\u2019re \u201cfree of copyright claims.\u201d (The hedged language doesn\u2019t come as a massive surprise considering that the original Imagen<\/a> was trained on a public data set, LAION<\/a>, known to contain copyrighted works.)<\/span><\/p>\n \u201cWe have to make sure that we\u2019re completely within the balance of respecting all of the laws that pertain to copyright information,\u201d Bardoliwalla continued. \u201cWe\u2019re very clear with customers that we provide them with models that they can feel confident they can use in their work, and that they own the IP generated from their trained models in a completely secure fashion.\u201d<\/span><\/p>\n Owning the IP is another matter. In the U.S. at least, it isn\u2019t clear<\/a> whether AI-generated art is copyrightable.<\/p>\n One solution \u2014 not to the problem of ownership, per se, but to questions around copyrighted training data \u2014 is allowing artists to \u201copt out\u201d of AI training altogether. AI startup Spawning<\/a> is attempting to establish industry-wide standards and tools for opting out of generative AI tech. Adobe is pursuing<\/a> its own opt-out mechanisms and tooling. So is DeviantArt, which in November\u00a0launched<\/a> an HTML-tag-based protection to prohibit software robots from crawling pages for images.<\/p>\n Image Credits:<\/strong> Google<\/p>\n<\/div>\n Google doesn\u2019t offer an opt-out option. (To be fair, neither does one of its chief rivals, OpenAI.) Bardoliwalla didn\u2019t say whether this might change in the future, only that Google is \u201c<\/span>inordinately concerned\u201d with making sure that it trains models in a way that\u2019s \u201cethical and responsible.\u201d<\/p>\n That\u2019s a bit rich, I think, coming from a company that canceled<\/a> an outside AI ethics board, forced out\u00a0prominent<\/a> AI ethics researchers and is curtailing<\/a>\u00a0publishing AI research to \u201ccompete and keep knowledge in house.\u201d But interpret Bardoliwalla\u2019s words as you will.<\/span><\/p>\n I also asked Bardoliwalla <\/span>about steps Google\u2019s taking, if any, to limit the amount of toxic or biased content that Imagen creates \u2014 another problem with generative AI systems. Just recently, researchers at AI startup Hugging Face and Leipzig University published a tool<\/a> demonstrating that models like Stable Diffusion and OpenAI\u2019s DALL-E 2<\/a>\u00a0tend\u00a0to produce images of people that look white and male, especially when asked to depict people in positions of authority.<\/p>\n Bardoliwalla had a more detailed answer prepped for this question, claiming that every API call to Vertex-hosted generative models is evaluated for \u201csafety attributes\u201d including toxicity, violence and obscenity. Vertex scores models on these attributes and, for certain categories, blocks the response or gives customers the choice as to how to proceed, Bardoliwalla said.\u00a0<\/span><\/p>\n \u201cWe have a very good sense from our consumer properties of the type of content that may not be the kind of content that our customers are looking for these generative AI models to produce,\u201d he continued. \u201cThis<\/span> is an area of significant investment as well as market leadership for Google \u2014 for us to make sure that our customers are able to produce the results that they\u2019re looking for that doesn\u2019t harm or damage their brand value.\u201d<\/p>\n To that end, Google is launching reinforcement learning from human feedback (RLHF) as a managed service offering in Vertex, which it claims will help organizations maintain model performance over time and deploy safer \u2014 and measurably more accurate \u2014 models in production. RLHF, a popular technique in machine learning, trains a \u201creward model\u201d directly from human feedback, like asking contract workers to rate responses from an AI chatbot. It then uses this reward model to optimize a generative AI model along the lines of Imagen.<\/p>\n Image Credits:<\/strong> Google<\/p>\n<\/div>\n Bardoliwalla says that the amount of fine-tuning needed through<\/span> RLHF will depend on the scope of the problem a customer\u2019s trying to solve.\u00a0 There\u2019s debate within academia as to whether RLHF is always the right approach \u2014 AI startup Anthropic, for one, argues<\/a> that it isn\u2019t, in part because RLHF can entail hiring scores of low-paid contractors that are forced<\/a> to rate extremely toxic content. But Google feels differently.<\/p>\n \u201cWith our RLHF service, a customer can choose a modality and the model and then rate responses that come from the model,\u201d Bardoliwalla said. \u201cOnce they\u00a0<\/span>submit those responses to the reinforcement learning service, it tunes the model to generate better responses that are aligned with \u2026 what an organization is looking for.\u201d<\/p>\n Beyond Imagen, several other generative AI models are now available to select Vertex customers, Google announced today: Codey and Chirp.<\/p>\n Codey, Google\u2019s answer to GitHub\u2019s Copilot, can generate code in over 20 languages including Go, Java, Javascript, Python and Typescript. Codey can suggest the next few lines based on the context of code entered into a prompt or, like OpenAI\u2019s ChatGPT<\/a>, the model can answer questions about debugging, documentation and high-level coding concepts.<\/p>\n Image Credits:<\/strong> Google<\/p>\n<\/div>\n As for Chirp, it\u2019s a speech model trained \u201cmillions\u201d of hours of audio that supports more than 100 languages and can be used to caption videos, offer voice assistance and generally power a range of speech tasks and apps.<\/p>\n In a related announcement at I\/O, Google launched the Embeddings API for Vertex in preview, which can convert text and image data into representations called vectors that map specific semantic relationships. Google says that it\u2019ll be used to build semantic search and text classification functionality like Q&A chatbots based on an organization\u2019s data, sentiment analysis and anomaly detection.<\/p>\n Codey, Imagen, the Embeddings API for images and RLHF are available in Vertex AI to \u201ctrusted testers,\u201d Google says. Chirp, the Embeddings API and Generative AI Studio, a suite for interacting with and deploying AI models, meanwhile, are accessible in preview in Vertex to anyone with a Google Cloud account.<\/p>\nImagen in Vertex<\/h2>\n
![\"Google](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/05\/ezgif-3-4146df7c01.gif?resize=800%2C500&ssl=1\")
<\/p>\n![\"Google](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/05\/Screenshot-2023-05-09-at-4.40.31-PM.png?resize=1024%2C576&ssl=1\")
<\/p>\n![\"Google](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/05\/ezgif-3-3f5ea18aaa.gif?resize=800%2C500&ssl=1\")
<\/p>\nNew models and tools<\/h2>\n
![\"Google](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/05\/image.gif?resize=720%2C405&ssl=1\")
<\/p>\n
![\"Read](\"https:\/\/i0.wp.com\/techcrunch.com\/wp-content\/uploads\/2023\/05\/google-io-2023-banner.jpg?w=1140&ssl=1\")
<\/a><\/p>\n<\/p><\/div>\n