5 ideas for your own AI grift with ChatGPT

AI entrepreneurship is being framed as a “gold rush” moment: the fastest path to profit isn’t inventing a new foundation model, but building narrow, useful applications on top of existing AI systems—then selling them to the right customers before the idea gets commoditized.

The pitch starts with OpenAI’s stack as the most accessible “shovel.” ChatGPT runs on an autoregressive language model lineage (GPT-3.5 is referenced), and the transcript points to GPT-4 as the next step, emphasizing the scale of training parameters as a proxy for capability. The practical takeaway is that developers don’t need to compete head-to-head with OpenAI; they can integrate the models through paid APIs. One example is “video-to-blog”: take a video, transcribe it (via OpenAI Whisper), generate structured headers with GPT, and optionally create accompanying images using Dolly. The underlying business logic is straightforward—turn a high-friction input (video) into a ready-to-publish asset (blog post) with minimal human effort.

The transcript then argues that “GPT isn’t the only game.” Alternatives include DeepMind’s language models (with Chinchilla and Alpha Code cited as examples of systems that have performed well in coding contexts) and open-source options like Bloom. For teams with data and technical depth, the route is training and deploying custom predictive models using MLOps tooling. Google Cloud’s Vertex AI is named as an end-to-end platform for ingesting data, training and evaluating models, and deploying them to serverless infrastructure. Hugging Face is also mentioned as an MLOps service offering many pre-trained open-source models as starting points.

From there, the emphasis shifts to specialized, high-value use cases where AI can feel magical to end users. Play.ht is highlighted for custom AI voice generation (including a claim about a fake podcast between Joe Rogan and Steve Jobs), while Beatoven is used as an example of generative AI music built from real musical inputs. Resume Worded is presented as an AI-assisted resume and LinkedIn analyzer that reduces the burden of writing and editing. The transcript also nods to both sides of the education market—tools that help students cheat and tools that attempt to prevent AI-assisted cheating—suggesting that demand will exist on multiple angles.

Business strategy gets its own section. One approach is to find a single painful problem and ship a focused app that can go viral—Restor Photos is cited as a Next.js app that restores grainy images to higher resolution and uses Replicate with an open-source model (gfp Gan) for the transformation. Another approach is to “AI-label” existing business models to make them look disruptive, even if the core mechanics are familiar—an AI lending platform example is used to argue that regression-based scoring can be enough to market as “AI,” with the transcript referencing Upstart’s IPO narrative. Finally, it pushes enterprise and government sales as the route to larger money.

The closing section turns cautionary: AI ideas can become obsolete quickly, big tech can dominate due to data and compute advantages, and scammers will increasingly exploit generative tools. Voice cloning, deepfakes, and AI-assisted scam writing are described as escalating threats. Despite that, the transcript ends on an intentionally absurd note about “uploading” an artificial persona into a robotic doll—positioning it as a speculative but plausible direction for future business models.