Generative AI represents a new set of expectations for our relationship with software to be more like our relationships with each other. With ChatGPT and GPT-4, all we need to do is type a message, like a text to a friend, and complex human-like text, images or computer code, can be served up on command.
Before the public launch of ChatGPT at the start of the year, when our relationship to artificial intelligence became integral to public discourse and dinner party conversations around the world, much to many dry January guests’ dismay– ten years ago, our art started to reckon with the same ideas. Ex Machina and Spike Jonze’s Her, which came out in 2013, explored how the intrusion of technology into every aspect of our lives would change our behaviors (from falling in love, to walking around talking aloud to computers), and legitimately change the way we comprehend what it means to be human.
Around that same time, we began to invest in the very founders who were at the forefront of developing these capabilities. It was well before “Generative AI” was a buzzword, but just under ten years ago, we backed Dennis Moretensen at x.ai, whose AI-driven personal assistant let people schedule meetings using plain English and nothing more than a cc’ to a chatbot, as well as Drew Silverstein at Amper Music, an AI composer, performer, and producer that creates unique music, tailored to any content, instantly.* These early forays democratized the ability for anyone to enjoy the benefits of a personal assistant for scheduling, or the ability to express oneself musically without years of technical training.
These companies were ahead of their time, and represented a relationship to technology we were, and still are, bullish about. Where our interaction with software is just as intuitive as collaborating with another person. Where software moves beyond enabling efficiencies to actually inspiring creativity and new, novel discoveries.
Today, there are a number of founders in our portfolio of 100+ data science and AI-enabled startups who are experimenting with Generative AI. Here, five of our founders share examples of how they are applying Gen AI in their businesses, and the near-term use cases they are most excited about.
Facet’s Joseph Reisinger on how “integrative” intelligence and coordinating action at scale will yield an era of creative megaprojects
AI-powered creation has been our core thesis at Facet since day one. The current wave of generative models allows us to move beyond photo editing and expand Facet into a full stack solution for visual storytelling. Our users are increasingly looking to move their workflows from photography to synthography—dialing in exactly the exact image they need in context, under budget.
With new technology, the nature of creative work is changing rapidly, shifting the balance of power from the professional adept class of creators to highly-motivated, entrepreneurially minded storytellers who need to do more with less. Generative AI products will gain the most traction empowering such entrepreneurially-minded professionals / “communicators,” freeing up their time to solve problems and removing one link in the production chain. Suddenly you don’t need a content team, or a creative team, or even to read the textbook to do your homework.
These days, I’m thinking a lot about “integrative” intelligence, or Englebart’s “collective IQ.” The internet is unlocking ad-hoc and anarchic collaboration at scales never seen before. Individual intelligence, IQ tests, whether or not ChatGPT can pass the bar exam, etc., all seem to me to be red-herrings.
What matters—maybe the only thing that matters—over the next 50, 100 years is our ability to coordinate action at scale. Coordinate with each other—and, increasingly—coordinate with our AI partners. In that vein, I expect the next 50 years to be an era of creative megaprojects.
Just like GitHub unlocks collaboration at scales only previously reachable by the most well-capitalized enterprises, Facet will bring artists, developers, storytellers and entrepreneurs together in ways that were simply impossible before.
Hyper’s Aaron Ng on constraints shifting from what people can find or make, to simply the limits of their creativity
As an avatar identity and storytelling company, a lot of what we’re constrained by is how much art people have tools to use. We’ve been testing generative 2D assets as a way to give people more storytelling tools. It’s exciting to us because it continues to lower barriers for our own production pipeline by increasing the number of assets people have for storytelling. A week-long process was squeezed into a single day.
Generative AI will dramatically decrease production time for everything. However, taste levels will rapidly rise due to the lower barrier-to-entry and level of rapid iteration that it unlocks (which is where the taste developed by senior artists will shine). We’ll see a lot more better things, faster, everywhere.
We’re excited about generative assets for characters, environments, and more. As generative AI improves, we’ll start seeing an infinite variety of assets (both 2D and 3D) that people can then use to tell exciting stories with their characters. The constraint will no longer be on what people can find or make but the limits of their creativity.
Text Blaze’s Dan Barak on how the race is on for the most up-to-date, contextual, and personalized responses anywhere
Text Blaze helps users streamline work anywhere by creating their own slash commands for any app – smart templates that can be used anywhere with keyboard shortcuts. Generative AI is one of the building blocks that our users can use to build their smart templates, allowing them to easily create and trigger contextual prompts anywhere. For example, draft a personalized introduction note on LinkedIn, or draft a response to a customer question that incorporates information from the CRM.
Inherently, these advancements will change how we work and how we approach our communications. Some entry-level communication-centric roles, such as customer support, recruiting and operations (and even engineering and design), will be automated or executed by lower-skill employees using Gen AI. This will require generating up-to-date, contextual and personalized responses, which is where we’ll see a lot of innovation. SEO content writing will shift to focus on Gen AI-optimization–competing to be the response generated for top prompts.
Kasisto’s Zor Gorelov on the importance of democratized financial knowledge and tools that ensure ethical, compliant AI
Kasisto has always existed with the purpose of democratizing financial knowledge. Our KAI platform helps bankers and banking consumers to access accurate, highly personalized financial knowledge in a secure, conversational environment that is directly connected to their banking ecosystem. To this end, we have been using Large Language Models (LLMs) since 2018 when we first trained a BERT-based model on vast amounts of conversational data collected from KAI’s conversations with users around the world.. This model, called KBERT, is trained to specialize in banking and financial services use cases.
Since GPT-3 became available, we have also incorporated it into our software stack and, most recently, used it to create personalized answers within our KAI platform. We are excited by the initial results, which show a noticeable improvement in the comprehension of our generated answers. We are continuously exploring many different approaches to incorporating LLMs into our stack, including the notion of a Private AI that is trained for our banking clients using their proprietary, secure documents and knowledge management libraries.
We are currently focused on the following Generative AI use cases, which we explore in greater detail in our recent conversation with Kasisto’s CTO, Sasha Caskey:
- Private AI: This technology fine-tunes LLMs on proprietary data sets so that the system can accurately answer very specific questions from these proprietary knowledge bases. There are a wide range of applications for this approach, which enables an institution to focus users directly toward a specific corpus of data.
- Personalized, Connected Assistance: This approach uses LLMs to automatically and securely access APIs from a variety of private financial data sources, so that the response system can tap directly into authenticated, user-specific banking information when generating responses. This gives users the ability to drill deeply into their own information and receive highly personalized financial insights.
- Conversational Orchestration: We’re producing technology that allows companies to safely and securely deploy and manage an entire team of inter-connected virtual assistants, bridging across both public and private knowledge sources. This gate-keeping technology ensures that users always have access to the most comprehensive sources of information while still receiving the most accurate, secure, and appropriate answers.
The Kasisto team is extremely excited about the prospects for Generative AI, and we’re energized by the amount of interest that is emerging in the marketplace. The technology advancements will fuel a new era of innovations and products in the field. We also predict the emergence of secondary markets for tools that help to ensure the AI will function at its best (complying with regulations, eliminating bias, ensuring truthfulness, etc.)
Ultimately, we believe this new wave enables us to better fulfill our purpose…of helping consumers to better understand complex financial products and make better decisions, and helping financial institutions to create more relevant, accessible products for their customers.
Osmo’s Alex Wiltschko on the unwritten future of cognitive automation, through the lens of fragrance molecule creation
The design process for making new aroma ingredients today is entirely manual. A fragrance chemist starts with a molecule that they know smells great, modifies it slightly, and tests it. Teams of fragrance chemists might do this thousands of times per year, and find just one or two that satisfy all of the requirements I mentioned above, making it good enough to put on the market.
With Generative AI, we’re able to enable our chemists and perfumers to explore and evaluate billions of novel aroma ingredients in seconds, and design them to precise specifications of quality and safety. The ingredients we’ve made so far smell great in the hands of master perfumers, industry experts, and double-blind odor panels.
Generative AI, and cognitive automation in general, is shortening the design-build-test cycle across industries. Osmo is committed to applying the absolute best tools to unlock new forms of creativity in the fragrance industry.
The future around cognitive automation and generative AI is unwritten, and I expect to be surprised by new tools for artistic expression, and even new kinds of art enabled by these rapidly maturing tools.