I tried to create ChatGPT in 1999 – this is what failure taught me
Fred Plais created a natural language search engine in the early days of the internet. Here’s what the AI founders of today can learn from his mistakes
How did you feel when ChatGPT was released in 2022? Amazed by the advances of technology? Intrigued by the possibilities it created? Maybe worried about what AI could replace?
Me? I could barely speak or sleep. ChatGPT (and other LLMs) did what we tried to do with a company I founded in 1999, which was killed by the dot-com crash.
But with that failure came a lot of lessons – both for me, and today’s AI start-ups.
ChatGPT before AI
I co-founded search engine, Infoclic, in 1999. Google was yet to go mainstream, and many of the search leaders were relatively primitive. Relevancy of searches was remarkably low.
It was also a time of massive growth, when the online world was opening up to people who weren’t tech-savvy. A “natural language” search engine, where you could ask questions rather than focus on keywords, made sense.
We built the first version of Infoclic in six months and continued to develop it in partnership with content providers. Traffic was growing and the business model, based on advertising, was sound, with expectations to break even.
But then the bubble burst.
Ad rates plummeted, and Infoclic was simply no longer viable. Google survived as the default search engine because of its smarter approach: ranking websites by popularity, based on incoming link, and searching using keywords.
All of which made ChatGPT’s launch very difficult. It was what I wanted to build in 1999, but it just wasn’t possible. More than two decades of technological advances, plus access to masses of data and processing power means that LLMs can process natural language and give people exactly what they need.
Lessons on timing and resilience
It would be easy to be bitter about another business realising your dreams, albeit more than two decades later. But back then, Google was the superior product and would have always caused the end for Infoclic, even without the dot-com crash. Our approach was wrong for the time – the technology simply wasn’t there yet.
Of course, start-ups only work if there’s a good idea that works and a quality team behind it. But the last piece of the jigsaw, and arguably the most important, is timing – which needs a great deal of luck and a little bit of magic. The business world is full of disasters caused by a failure to act quickly enough, such as Blockbuster and Kodak, and innovations that were too far ahead of their time, like the Nintendo Virtual Boy and, of course, Infoclic.
Founding a business is like taking an accelerated MBA. You learn how to hire a team, convince investors of your vision, build a product, run an engineering team, and provide customer service. You need to quickly become an expert in everything a business needs.
Like the dot-com bubble, there’s undoubtedly a similar story for AI today. Start-ups are attracting crazy valuations and levels of investment, and we will see at least one correction to this. From experience, I would advise founders of AI start-ups to be mindful of these valuations. They might seem promising today, but could jeopardise the future of your business.
The dot-com crash didn’t kill the internet, and any AI correction won’t kill AI
Some of the start-ups that will fail will have been merely riding the wave of investment, doomed to failure, but we will also see the demise of promising businesses that have good ideas, good people, and unfortunate timing.
The most valuable lesson these entrepreneurs will learn is resilience. The dot-com crash didn’t kill the internet, and any AI correction won’t kill AI. But it will create a new cohort of entrepreneurs that have learned valuable lessons, ready to go again when the time is right.
Fred Plais is CEO of