Yes, your AI startup sucks
Probably more than the web3 platform you launched two years ago
With the release of ChatGPT and the availability of the OpenAI API, thousands of startups emerge with the promise of replacing <insert-a-job-title-here> or <make-you-100-times-more-productive-at>.
Almost all startups are rushing to integrate AI into their products. And investors are urging their portfolio companies to add AI to increase the company valuation. If almost all startups created in 2021-2022 had a web3 or crypto component, most startups from 2023 have an AI one.
No one can improvise being an AI expert. And many grifters label themselves as AI expert when they barely understand linear algebra. There is a true scarcity of engineers who are experts in machine learning. The few good and hard-working ML engineers are already offered massive amounts of money. Many startups are left with engineers who can build (or rather integrate) a good AI feature but not a great one. As a result, they end up with marketing stunts rather than real engineering work. This tweet summarizes what is happening today.
Creating a dataset, building and training a model are huge endeavors that take time (it took months to train GPT3 on 45 terabytes of data). And many AI-based products or features are not original work done by a company but just using the OpenAI API. New AI-based products are not original but a proxy to use the Open AI API.
Using the same platform and model for all these new products will make them look the same, without a clear differentiator. In addition, overfocusing on integrating AI (and not being an expert) diverts focus from core aspects (design or user experience).
Adding a great AI feature can be the cherry on top of your product. But a bad AI can destroy the user experience and drive people away. Remember when Microsoft introduced clippy in Office? Clippy was supposed to help users and make them more productive. But the product was so bad that Microsoft was forced to remove the assistant from their product.
Many 2023 AI startups want to look like ChatGPT but look more like clippy.
Do not get me wrong: the progress in AI is impressive. I am convinced this is just the beginning. It will take time to see new companies emerge and new models built. I am certain we will see breakthroughs in the next five years, and I can’t wait to see innovative AI-based products. But it will take time, and just adding a few API calls to an AI service will not make a great product.
The main players will not change. OpenAI will keep pushing the fantastic work they are doing. Companies like Hugging Face will keep attracting talented engineers. And the next big shots will take months or years to build.
Building a great company and an awesome product takes time. Nobody talked about OpenAI 5 years ago. It took them 6 years to get significant traction. And seven years to release a hit product. Nothing replaces hard work, especially not a few API calls.