VP of Generative AI at Adobe Alexandru Costin on The Future of Content Creation
Ep. 16, Unsupervised Learning
In this episode of Unsupervised Learning we sat down with Alexandru Costin, VP of Generative AI and Sensei at Adobe and former head of Adobe Romania, for an exciting conversation on one of the few enterprises capitalizing off new generative AI methods. See the full episode below!
Alex provides insights into Adobe’s early beginnings in generative AI pre-ChatGPT when generative adversarial networks (GANs) became popular in 2019, Adobe’s technology transfer pipeline that helps them commercialize cutting edge AI research, and some of the considerations beyond the models such as de-biasing training data and guardrails to ensure consistent output. You can listen to the full conversation on Spotify, Apple, and YouTube or read our highlights below!
⚡️ Highlight 1: Creators could become 10x more productive with Generative AI
One of the most controversial topics since the release of products like Stability.ai’s Stable Diffusion and OpenAI’s ChatGPT has been the fear of job loss due to generative AI. To be fair, these concerns are warranted as global news start up Rest of World, funded by Sophie Schmidt (daughter of Google founder Eric Schmidt), reported the number of game illustrator jobs in China significantly decreased in part due to new AI tools. AI is also a hot topic in SAG-AFTRA’s writing strike. However, these concerns have not necessarily come to fruition based on Adobe’s experience.
“Throughout each of those [previous technological] waves, the innovation, we have seen the same pattern apply where content creation gets cheaper and easier to do and more accessible to a larger audience, but which triggers the concerns for the creators about the role of their job security, the role they will play in this new world. With every transition, what happens is the demand for digital content or for content in general increases exponentially more than the decrease in cost of creating content. So because of that, we have actually seen for every transition an opportunity for our customers that we serve through various product lines to become relevant, stay relevant in this new world, and actually embrace the new technology, not be disrupted by it, and become 10X more productive or creative.”
While there are many concerns on the negative impact of generative AI, there are just as many, if not more, reasons to be excited about the possibilities for content creators in fields like marketing. Many startups like Writer and Jasper have already made waves in this space.
⚡️ Highlight 2: Creating the generative AI user experience
ChatGPT democratized access to AI for anyone who knew how to type commands into a keyboard. However, is language the end all be all for human-computer interaction with generative AI products? Alex provided his thoughts on the topic in this week’s episode.
“So when we think about where those user experiences would land, we think it's going to be a mix of language with pointing and sliders. I don't think we've reached the end state of where the human computer interaction for creativity will reach, because we do think when you just type something normally most of the time you don't get what's in your mind, you get something else. So there's some frustration. I have a daughter, she's 11, Rebecca. The first time I showed her Firefly, she started typing in things and she said, "Yeah, this is not what I imagined. It's not working dad." But I couldn't know what's in her head when she was typing pink rabbit or whatever the particular scene was at the time.”
Check out Adobe Firefly, a family of generative AI products designed for creators, including the famous generative fill! Another interesting takeaway here was that generative fill uses the rest of the image as an implied prompt to ensure that the generated content fits appropriately. Adobe also has their Sensei GenAI platform that serves as a co-pilot for marketers and customer experience managers.
⚡️ Highlight 3: Making generative AI safe for the enterprise
One of the main blockers for enterprises using generative AI is having the ability to control the content created, and ensuring it’s aligned with company policies. Start-ups like Guardrails are creating libraries for developers to implement custom rules to ensure consistent output for LLMs, meanwhile Adobe is working to create guardrails for AI generated images.
“[Enterprise customers] want to make sure that content created either respects the company's brand and/or if you're in a particular campaign and you generate additional images, they have to respect the campaign brand because they might have sub-brands. And this is where actually generative AI can play a key role, and we're investing in innovating in this space. How can we enable enterprises or the marketers or the creative department inside an enterprise create content that is automatically compliant with the brand so you don't have to QA it a lot and it's beautiful out of the gate.”
In addition to guardrails that reduce QA time after content generation. It’s equally important to ensure the AI-generated content is unbiased and not trained on content that puts their enterprise customers at risk of copyright infringement. There are countless examples of models released by generative AI providers perpetuating human-bias in AI-generated content, and several copyrights lawsuits from human artists. Adobe found a solution to both of these risks through training their models on Adobe stock photos and leveraging AI to de-bias their training data.
“We have a model that helps us understand if there's a person reference in a prompt. And most of the time when you refer a person, you might refer it through a job, you could say a lawyer, and in a biased model, when you put in lawyer, you will get Caucasian in their sixties and males…so what we did is we detect people, we detect jobs, and we de-bias the actual content to introduce the right distribution of skin tones and the right distribution of genders and age groups from the country of origin of the request.”