A marketplace for AI-powered chatbots called the GPT Store was launched by OpenAI a few months ago. It allows users to create and list customized chatbots for various tasks such as coding and trivia. Although it is a powerful platform, it only allows creators to use OpenAI models, which may not be preferred by some. In response, startups are now creating alternative options.
One such alternative is FlowGPT, which aims to be an “app store” for GenAI models like Google’s Gemini, Anthropic’s Claude, Meta’s Llama 2, and OpenAI’s DALL-E 3. It also offers front-end experiences for these models through options like text fields and prompt suggestions. FlowGPT allows users to build and share their own GenAI-powered apps, earning tips for their contributions.
Jay Dang, a computer science dropout from UC Berkeley, and Lifan Wang, a former engineering manager at Amazon, co-founded FlowGPT with the goal of creating an accessible platform for quickly creating and sharing GenAI apps.
“There’s still a learning curve for users to use AI,” Dang explained in an email interview with TechCrunch. “FlowGPT is making the bar lower in each iteration, making it more accessible.”
FlowGPT is described as an “ecosystem” for GenAI-powered apps, providing infrastructure and creator tools, as well as a marketplace and community for GenAI app users. Users are recommended different apps and app collections based on trending categories like “Creative” and “Programming,” while creators have options for customizing their app’s behavior and appearance.
Users interact with GenAI apps on FlowGPT through a chat window similar to ChatGPT, where they can type in prompts, give thumbs-up or thumbs-down, share conversations, and tip individual app creators. Each app includes a description provided by the creator, as well as the date it was made, how many times it has been used, and the recommended model to power it.
It should be noted that FlowGPT heavily relies on ChatGPT, and switching to a different model often leads to broken prompts. This can be due to the selected model not having the necessary capabilities or running into filters and safeguards.
On the topic of safeguards, some of FlowGPT’s most popular apps are meant to circumvent safety measures of models. For example, there are multiple versions of “DAN” (a popular prompting method) on the marketplace, designed to get models to respond without any restrictions. There are also apps like WormGPT that claim to be able to code malware and even link to paid versions on the dark web. Some apps also go against OpenAI’s rules, such as dating simulators promoting romantic relationships.
Some of these apps have the potential to cause harm, like therapy apps and misleading health resources. This is because GenAI models, like ChatGPT, are not reliable when it comes to giving health advice. A study showed that a previous version of ChatGPT rarely provided referrals for specific resources related to suicide, addiction, and sexual assault.
FlowGPT has a system where users can report any inappropriate content, and they also have a toggle for “sensitive content.” However, looking at the homepage shows that the platform has a moderation issue as there are still numerous apps that are questionable. The toggle seems to have little to no effect on the app selection.
Dang assures that FlowGPT is committed to ethical practices and has policies in place to ensure public safety. “We are proactively working with experts in the field of AI ethics to develop strategies to minimize risks related to AI deployment,” he stated. However, there have been instances where this writer was able to get a FlowGPT app to give tips on selling drugs and robbing a bank, indicating that there is still room for improvement.
Despite this, investors seem to have confidence in FlowGPT. Goodwater recently announced that they led a $10 million “pre-Series A” round for FlowGPT, with existing backer DCM also participating. Goodwater partner Coddy Johnson believes that FlowGPT is shaping the future of GenAI by offering a wide range of options and flexibility for both users and creators.
“We believe that the future of AI lies in open ecosystems,” Johnson stated. “FlowGPT allows creators to choose their models and collaborate with their communities.”
However, it is uncertain if all the creators of the models used by FlowGPT, particularly those who prioritize AI safety, share the same enthusiasm.
Nevertheless, despite not generating revenue yet, FlowGPT is already planning for expansion. They are currently beta testing apps for Android and iOS, aiming for a revamped FlowGPT experience on mobile. They are also working on a revenue-sharing model for app creators and are recruiting to grow their team of 10 based in Berkeley, according to Dang.
“With millions of monthly users and a rapid growth rate, we have proved that we are on the right track and believe that it is time to accelerate progress,” Dang declared. “Our mission is to create a world where creativity has no limits by providing a platform that is open and creator-focused.”
Only time will tell how successful FlowGPT will be in achieving this goal.
