FlowGPT is the Wild West of GenAI apps

Just a few months ago, OpenAI launched the GPT Store, a marketplace where people can create and list AI-powered chatbots tailored to perform a variety of tasks (e.g. coding, answering quizzes, etc.). The GPT Store is undoubtedly powerful. However, usage requires the usage of OpenAI models and never others, which some chatbot creators – and users – object to.

So startups create alternatives.

One, FlowGPTgoals to be a type of “app store” for GenAI models like Google’s Gemini, Anthropic’s Claude, Meta’s Llama 2, and OpenAI’s DALL-E 3, in addition to front-end experiences for these models (think text fields and prompts). Through FlowGPT, users can create their very own GenAI-based apps and make them publicly available to receive suggestions for his or her contributions.

Jay Dang, a UC Berkeley computer science dropout, and Lifan Wang, a former engineering manager at Amazon, founded FlowGPT last 12 months out of a shared desire to create a platform where people could quickly construct and share GenAI apps.

“There continues to be a learning curve for users to make use of AI,” Dang told TechCrunch in an email interview. “FlowGPT lowers the bar with each iteration, making it more accessible.”

Dang describes FlowGPT as an “ecosystem” for GenAI-based apps – a group of infrastructure and constructing tools tied to a marketplace and community of GenAI app users. Users receive a feed of apps and app collections really useful to them based on trending categories (e.g. Creative, Programming, Gaming, Academic), while developers have options to customize behavior and appearance – obtained from GenAI apps.

Users interact with GenAI apps on FlowGPT through a chat window not unlike ChatGPT, with options to enter prompts, thumbs up (or thumbs down) apps, share links to conversations, or share suggestions for people App creator. Each app has an outline provided by the creator, indicating the date it was created, how continuously it’s used, and the model the creator recommends for operation.

I say this because, at their core, FlowGPT apps are literally prompts – prompts that encourage models to reply in certain ways. For example, the “Scared Girl from horror filmThe app instructs ChatGPT to inform – because the title suggests – a horror story a few scared girl. “TitleTuner“causes ChatGPT to optimize headlines so that they rank higher in engines like google. And SchoolGPT uses ChatGPT for step-by-step solutions to math, physics and chemistry problems.

You will notice the heavy reliance on ChatGPT. If you utilize FlowGPT long enough, you may also notice that many prompts stop appearing when the model changes from the default.

Sometimes it’s because the chosen model doesn’t have the correct capabilities. In other cases, the prompt encounters a model’s filters and protections.

On the topic of protective measures…

Some of FlowGPT’s hottest apps are essentially jailbreaks designed to bypass models’ security measures. There are several versions of DAN available on the market – “DAN” is a well-liked prompting method used to make models reply to prompts without being sure by their usual rules. Elsewhere, there are apps like WormGPT that claim to give you the option to program malware (and link to paid dark web versions of the chatbot that may do more), and dating simulators that run afoul of OpenAI Rules against encouraging romantic companionship.

Many of those apps could potentially cause harm, akin to therapy apps and apps that promote themselves as trusted health resources. GenAI models like ChatGPT are notoriously poor health advisors study shows that an earlier version of ChatGPT rarely Providing referrals to specific resources for help related to suicide, addiction, and sexual assault.

Any app on FlowGPT that’s objectionable – akin to giving instructions on generate deepfake nude photos using an AI image generator (and there are several that do that) – may be reported to the platform’s community manager for review. And FlowGPT offers a “sensitive content” switch.

Photo credit: FlowGPT

However, a take a look at the homepage clearly shows that FlowGPT has a moderation problem. It’s the Wild West of GenAI apps – and the switch is so ineffective that I barely notice a difference in app selection when it’s on.

Dang swears up and down that FlowGPT is indeed an ethical and compliant platform, with risk mitigation policies aimed toward “ensuring public safety.”

“We are proactively working with leading experts in the sector of AI ethics,” he said. “Our collaboration focuses on developing comprehensive strategies to attenuate the risks related to AI use.”

Considering this creator received a FlowGPT app that provides instructions on sell drugs and rob a bank, I’d say the corporate still has some work to do.

Investors appear to disagree.

This week, Goodwater announced that it led a $10 million “Pre-Series A” round in FlowGPT with participation from existing backer DCM. Goodwater partner Coddy Johnson said in an email conversation with TechCrunch that he believes FlowGPT “helps prepared the ground” in GenAI by giving each users and developers “probably the most alternative” and “probably the most flexibility and freedom”.

“We imagine the best way forward for AI lies in open ecosystems,” Johnson added. “FlowGPT (allows) creators to decide on their models and collaborate with their communities.”

Photo credit: FlowGPT

I’m not so sure that every one maintainers of the models utilized by FlowGPT – especially those committed to creating AI safety a top priority – will share this enthusiasm.

Nevertheless, and in lieu of the impact of those providers (at the least on the time of publication), FlowGPT – which isn’t yet generating revenue – is laying the groundwork for expansion. The company is testing beta apps for Android and iOS that may bring a revamped FlowGPT experience to mobile devices, working on a revenue share model for app developers and recruiting to expand its Berkely-based team of 10, Dang said.

“With hundreds of thousands of monthly users and a rapid growth rate, we’ve already proven that we’re on the correct track and we imagine it’s time to speed up progress,” he continued. “We are setting a brand new standard for immersion
AI-driven environments that provide a world where creativity has no limits… (Our mission stays to create a more open and creator-focused platform.”

We’ll see how far this goes.