Our team attended the BrXnd.AI conference in NYC earlier this month. The room was filled with everyone from AI enthusiasts and those who are slightly terrified at what's to come. After a day of listening to humble experts, the quote that kicked off the day never seemed more true. “If you are just writing SEO headlines with ChatGPT you are already lost.” We would say we were left questioning everything we knew, but we didn’t even know where to begin with our questions. The main takeaway we had is that the possibilities are infinite.
BUILDING INTUITION IN AI
The first session, Building Intuition in AI, was led by Noah Brier, founder of BrXnd.AI and the host of the event. AI is counterintuitive. It goes against how we typically tend to operate. When trying to use a platform's API (Application Program Interface) to create a plugin the platform usually tells you what data they need in exchange for certain data. Basically, you give us this data and this is what we will give you. ChatGPT essentially has you tell them what data you will give them and what you want in return. Counterintuitive. To help paint the picture, Brier compared it to going to Ikea, getting a box of every part they ever made, you sketched out what you wanted, and then Ikea gave you instructions on how to build that specific thing. This analogy filled the room with a synchronous “ahhhhh.” Brier went on to continue making the point that it runs counter to our existing intuition and we need to build it back up.
THE COLLABS YOU’VE ALWAYS WANTED
One of Brier's experiments is called CollXbs. It’s an AI site that allows you to create a mashup of two brands and an item of your choice. While some of the CollXbs that were shared looked like it would need some work before it went to market, a lot of them were pretty legitimate. What he found was that good brands make better images. One of the more impressive takeaways was that AI models hold a similar sense of what a good brand is as we do. You could even tell which brand AI thought was stronger based on what brand is overpowering in the CollXb. They didn’t even need to be a brand like Coca-Cola or Nike. It picks up on brands like Aime Leon Dore and Warby Parker. CollXb was able to understand the Supreme product look & feel but also it’s art direction. When it gave a Supreme x Ralph Lauren sweater, the image was shot as though it was the Creative Director of the brand behind the camera. Sometimes the category triumphed over the brand. AI was asked to create a Saucony shoe collab and it came out with a Swoosh. AI so closely associated shoes with Nike that the Swoosh was not a Nike element to the machine but rather just an element of shoe design. Brier got some of these products consumer tested by System1 and they came back in the 90th percentile of anything the brand has ever done. While I don’t think Peppa Pig and Chanel are going to partner up anytime soon it is interesting to see what a hoodie would look like.
THIS MACHINE IS REALLY BAD AT MATH
After Noah set the scene of the excitement to come throughout the day, the next session, Hallucinations! For Fun and Profit, led by Tim Huang was where all the pieces started to click. He showed an example of a 5th grade level math problem that ChatGPT got wrong. He also showed an example of someone asking ChatGPT for the weather with an answer of “75 and sunny”. He went on to explain that all ChatGPT does is predict what is the most likely phrase someone would say next. It uses and pulls data to understand how to best respond. So while the math answer was wrong, it was how someone would phrase the correct response. And while 75 and sunny was pretty close to the weather it didn’t actually check the weather. It just used data to understand that during May in NYC that is typically what the weather is like. ChatGPT predicts language and is not fact based. So LLMs (Large Language Models) are bad at everything you think they are good at (math, science, etc.) and good at everything you think they are bad at (creativity, storytelling, art, etc.). LLMs are concept retrieval systems, not fact retrieval systems. It understood the concept of that math problem but there was no actual calculation happening. AI is learning a vibe and a concept. They are making predictions.
THE POWER OF MAKING MACHINES HALLUCINATE
Once we collectively understood that these systems are great for those in the creative industry, Huang wanted to show us the power of getting the machines to hallucinate. Brand is a collective understanding of what a company represents to society. It is how people perceive you and not what lives in your internal strategy decks. LLM’s data on brands is external. It does not have access to your internal documents. It understands what you represent and what your audience resonates with you for. This allows hallucinations to be a feature rather than a bug. With language, you can make LLMs any interface you want. For example, you can tell it to be a Brand Translation Machine. You will give it a brand and text, and it will spit out that text and make it on-brand. Seeing the Gettysburg address edited into the tone of voice of coca-cola made everyone blink twice. “The sweet taste of freedom and equality.” One of the other hallucination examples Huang demonstrated was making the LLM a meter. It assigned objects a score based on the degree they align to a brand. Essentially giving a numerical value to how beneficial brand partnerships can be or how accurate consumer personas are. If you are Skims and are exploring a collaboration with Rimowa, the LLM will give you a numerical score you wouldn’t typically be able to calculate based on the alignment. Learning and creating new ways to manipulate these kinds of hallucinations will be one of the biggest benefits to those in the Ad and Marketing industry.
BRINGING THE AI BUZZWORD INTO YOUR CAMPAIGNS
Brands are starting to leverage AI for campaigns to generate buzz and capitalize on the moment. Some of these campaigns are not only relevant and timely because they incorporate AI but because they are innovative and are taking marketing to the next level. Paul Aaron who is the CEO of Addition took the stage and shared a little bit about what his agency is creating. They partnered with the New York Times to create an interactive AI experience that pulls headlines from New York Times articles to reflect who you are as a reader. Think of it as a Spotify Wrapped for the New York Times. You can try it out and read more about it here. Another moment that had everyone’s jaws dropped was a project called The Dreamkeeper. This platform uses an AI dream scientist to help you visually recall your dream and then record it. Using your description it creates an image of your dream and now can even create a video of your dream. Everyone’s dreams get preserved in a public gallery. Allowing you to travel in and out of your friends dreams. AI is really a dream. And if you are like us, and like to stalk the multi-million dollar homes for sale in the middle of the night this next one is for you. Addition partnered with Realtor.com to help you create and find your dream home. Using a text to image model, if you give it a description of your dream home (ie. a glass house on the beach or a converted school bus) it will create an AI image of your wildest imaginations. From there, it is integrated with Realtor.com and will actually show you houses similar to the AI generated image that are for sale and available. This just took house hunting to a whole new level.
YOUR CONSUMER PERSONAS ARE ALIVE
AI is coming to life. Or at least research is. Meet Person-AI, a chatbot manifestation of a brand’s consumer persona. Inspired by a chatbot's ability to use natural language, it sits on top of Chat GPT and makes it interactive. Ivan Kayser, CEO of Redscout, started this exploration with the idea of what if you could ask the question you forgot to ask? Here is how it works: You message the chatbot something like “describe your style in 5 sentences,” “if you were the CEO, what would you do?” or “what new category do you suggest we go into?” The chat bot then replies as your consumer personas in and feels in spirit like an actual customer. For every response it will even pull quotes from your customers as references for how it created your response. It’s like a chatbot focus group. Kayser reminded everyone at the conference that “your brand is how it exists in people's minds, not in how you're represented in your strategy decks.” Person-AI allows you to have direct communication with people and better understand how you are perceived by your consumers.
KEEP IT ON BRAND
As brands continue to tap into AI, they need to be mindful and aware of the content it is producing. Not only making sure there are no glitches but also confirming that it is brand aligned. Rob May, Founder & CEO of Nova Cloud, highlighted four key problems with generative AI: poor quality output filled with errors, outputs that aren’t on-brand, easy creation and use of brand by non-authorized users, and that humans can’t quality check at machine scale. As a test, he fed ChatGPT the brand guidelines for Tesla and asked for a few deliverables. The very first rule in the Tesla guidelines was “Don’t use the word luxury.” ChatGPT produced a sentence using rich and luxury as its first deliverable. Clearly not acting as a skilled brand marketer. May created BrandGuard for this reason. A tool that uses a digital style guide checks that anything you create digitally follows it. Essentially making machine readable brand style guides. It is a plug-in that connects to all platforms and gives you a score of how on-brand the content is. Allowing brands to create mass content using generative AI that always feels aligned.
CREATIVE WITH A MACHINE’S TOUCH
Diving into creative, there are so many touch points you can use AI to support in the process. The branding of the event itself was created by the agency Otherward using AI. Elliott Walker and Tim Hucklesby, Creative Directors and Co-Founders of Otherward, shared some learnings they noticed when creating the BrXnd.AI brand. The first being the speed and scale in which you can create imagery was unmatched. It required lots of tinkering but the output is much faster. Second was that no matter how good the prompts were, you are not in full control. You will never get the output of what you are envisioning in your head. Lastly, brands that are crystal clear in who they are had much more success in the AI programs understanding their brand. Once you have the brand you need to create content to maintain the brand. That’s where Flair comes in. Mickey Friedman, CEO & Co-Founder of Flair AI, gave a demo of how to use their platform for product photography. Flair is a composable design platform/widget that generates on brand, highly personalized photography. There were a few ways in which the platform creates an output of content. Design an asset from scratch by dragging and dropping props they have available within their system to create the set-up you want. Another way is to upload a new product image and a previous campaign image you have. The platform will generate a mash-up of the two allowing you to have new products in the same style of photography without a reshoot. Add a human touch with composable human models. Within the platform you are able to control the poses of the models and personalize the models to tailor them to your brand and to the individualized customer. Friedman predicted that Flair would be integrated on every e-commerce platform. To be available as a widget to be embedded into Instagram marketplace or Facebook. If you are unsure how to start using AI for creative needs, Jamie Robinson, Co-Founder and CCO of JOAN, uses it as a way to break the blank page. “Nothing is ever perfect coming out of it, but I am using it as a tool to give you a starting point. ‘That’s not right, but it gave me an idea.’” Next time you are lacking inspiration working from home, try using AI to help get the process started.
Copyright and Bias when it comes to AI
As future focused and eye opening as the BrXnd.AI conference was, we need to continue to have conversations on the legality and bias that comes with using these platforms. A panel called Copyright versus Generative Creativity, touched on the Drake & The Weeknd AI case. The reason the song was able to get taken down was because some of the lyrics featured in the song were previously copyrighted. Style of something is not copyrighted. Copyright viability is when it includes some component. It is important to know what the source of information the AI is using to prevent infringement. Copyright only protects human creativity. You can not copyright something purely AI created but you can copyright human manipulations to AI. Bias is something that also needs to be considered and kept top of mind as we integrate these platforms into our businesses and daily life. Meredith Broussard, Associate Professor at NYU, and Charlton Mcilwain, Vice Provost & Professor at NYU, had a conversation on how bias has a role in AI. They described AI as “really complicated, beautiful math.” AI uses data to inform and make decisions and it's important to understand where exactly that data is coming from. It can come from the US Patent office to Reddit to white supremacy websites. Automated systems discriminate by default. Platforms have put systems in places through real toxicity prompts data sets (essentially tells them what questions are harmful). Another point that was made by Broussard was that as users, we expect that the companies know what they are doing. It is important that we don’t trust everything that these platforms give us at the current state of development. Humans understand nuance and machines don’t.
All in all, the conference left us hungry to learn more, optimistic about the ways that it can help amplify the way we live and the way we work, and continually cautious how important it is to keep a human perspective in everything we do.
Other Articles
All ArticlesNo items found.
Future proof your brand.
Contact Us