OpenAI established the GPT Store a few months ago, allowing individuals to develop and list AI-powered chatbots that can code, answer trivia questions, and more. GPT Store power is real. It involves utilizing just OpenAI models, which some chatbot designers and users dislike.
Startups are offering alternatives.
FlowGPT aspires to provide an “app store” for GenAI models like Google’s Gemini, Anthropic’s Claude, Meta’s Llama 2, and OpenAI’s DALL-E 3 and their front-end experiences (text fields and prompt recommendations). FlowGPT lets people develop GenAI-powered applications and share them for tips.
Jay Dang, a UC Berkeley computer science dropout, and Lifan Wang, a former Amazon engineering manager, established FlowGPT last year to develop a platform for swiftly creating and sharing GenAI applications.
“There’s still a learning curve for users to use AI,” Dang emailed Eltrys. “FlowGPT lowers the bar with each iteration, making it more accessible.”
Dang calls FlowGPT an “ecosystem” for GenAI-powered applications, including infrastructure, creator tools, and a marketplace and community of users. Users receive a feed of applications and app collections suggested by popular categories (e.g., “Creative,” “Programming,” “Game,” “Academic”), while authors may customize GenAI app behavior and design.
FlowGPT’s GenAI chat window lets users enter questions, thumbs-up (or thumbs-down) applications, exchange links to chats, and tip app authors. Each app’s author provides a description, date, use, and recommended model.
I say model the developer suggests since FlowGPT applications are cues that prime models to react. The “Scared Girl from Horror Movie” software tells ChatGPT to narrate a horror narrative about a terrified girl. “TitleTuner” helps ChatGPT improve headlines for SEO. SchoolGPT uses ChatGPT to solve math, physics, and chemical problems step-by-step.
ChatGPT is widely used. Use FlowGPT long enough to discover that many prompts fail when the model is changed from default.
Sometimes the chosen model lacks the necessary capabilities. Sometimes the suggestion hits a model’s filters and defenses.
Regarding protections…
Some of FlowGPT’s most popular applications are jailbreaks that bypass model safety. “DAN” is a common prompting strategy used to induce models to react to requests beyond their typical rules. Other applications include WormGPT, which claims to code malware and links to paid, dark web versions of the chatbot, and dating simulators that violate OpenAI’s ban on romantic relationships.
Therapy applications and self-proclaimed health tools might hurt users. GenAI models like ChatGPT are notoriously lousy health advice-givers. One study found that an older version of ChatGPT seldom referred users to suicide, addiction, and sexual assault options.
FlowGPT’s community manager may assess any app that offends, such as one that teaches how to produce deepfake nudes using an AI picture generator. Additionally, FlowGPT has an option for “sensitive content.”
The homepage shows FlowGPT has a moderating issue. It’s the wild west of GenAI applications, and the setting scarcely affects app choices.
Dang insists that FlowGPT is ethical and follows the rules, with risk mitigation strategies to “ensure public safety.”
“We’re proactively engaging with leading AI ethics experts,” he stated. Our partnership focuses on establishing comprehensive AI deployment risk mitigation strategies.”
This writer obtained a FlowGPT app to deliver drug sales and bank robbery instructions; thus, the firm has to improve.
Apparently, investors disagree.
DCM joined Goodwater in a $10 million “pre-Series A” financing of FlowGPT this week. Goodwater partner Coddy Johnson told Eltrys via email that FlowGPT is “helping to lead the way” in GenAI by giving consumers and developers “the widest choice” and “the most flexibility and freedom.”.
“We believe open ecosystems are AI’s biggest future,” Johnson said. “FlowGPT lets creators choose their models and collaborate with communities.”
I doubt all the model maintainers of FlowGPT taps, especially those who’ve vowed to prioritize AI safety, would share that excitement.
However, FlowGPT, which isn’t yet profitable, is preparing for growth in the absence of vendor ramifications. Dang said the business is beta testing Android and iOS applications that would redesign FlowGPT on mobile, working on a revenue-sharing model for app producers, and hiring for its 10-person Berkely team.
He said, “With millions of monthly users and a fast growth rate, we’ve already proven we are on the right track, and we believe it’s time to accelerate the progress.” We redefine the benchmark for AI-driven settings, fostering limitless innovation. Our goal is to create an open, creator-focused platform.”
We’ll see how far that goes.
