AI’s viral moment

It’s been an interesting year or so to be an AI startup. Ever since the release of ChatGPT, followed by a very public AI arms race in 2023, specifically generative AI, and more specifically large language models (LLMs), went from being discussed in computer labs, startups, and tech companies, to being discussed at government policy meetings and families’ kitchen tables.

It’s no secret that OpenAI, the company behind ChatGPT, was itself somewhat baffled by its sudden popularity. In their view, its launch in late November 2022 was nothing more than the release of a more polished version of technology they had already released as InstructGPT and GPT-3.5 earlier in the year, and, before that, as GPT-3 all the way back in June 2020.

As to why ChatGPT proved to be a market-defining moment, a group of key OpenAI employees professed in a March 2023 MIT Technology Review, “honestly, we don’t understand. We don’t know.”

Of course, the OpenAI employees were somewhat underselling themselves: this was exactly what the company had intended to achieve with the launch of ChatGPT. It’s just that it had a tremendously outsized effect. The primary difference between ChatGPT and previous iterations is that OpenAI added some conversational data. Where previous versions were made available via API to software developers, i.e. a very specific audience familiar with the tech, ChatGPT focused on making the model accessible to the general public. And what had been a familiar technological capability by that point to specialised developers, baffled and excited the general public.

AI as a mature market

November 2022 was as much an end as it was a beginning. It was the moment when the proverbial “experiment escaped the lab.” As it roamed the streets, world governments started calling for summits and there was general shock and awe. The industry’s somewhat sheltered childhood was over, and its battle with regulators and public perception began in earnest.

It is a somewhat tired statement at this point, however – though no less true – that AI has been around for decades, and has been iteratively developed by hundreds of thousands of people and companies worldwide. My own company first started pitching AI-based solutions in the fashion, health, and AV control space back in 2017/2018, before I incorporated the current company in 2019.

In many ways, we were late to the game. When we participated in the well-known accelerator Techstars in 2020, about a third of our cohort were AI companies. Already then, I distinctly remember talk of an “AI bubble” and “AI hype” in conversations with venture capital. At the time, many “AI” companies in the market simply performed certain services manually, while seeking capital to build the machine learning that ultimately would automate those services.

One of the questions I get asked most often as an AI founder is if it is harder today, post-AI boom, to pitch an AI company. The hype, after all, means that the market is saturated: every new company seems to be an AI startup. In my experience, however, the opposite is true.

Back when I started pitching the current iteration of our company in 2020, AI was omnipresent but much less well-understood. During the due diligence process, venture capital investors often asked confused questions, and had difficulty differentiating between startups building radical new machine learning and those using AI as a buzzword. Since the hype, however, VCs have become extremely well-educated on AI. While companies that simply provide a wrapper around existing AI products – or those that essentially just have an idea for which they think AI can be deployed effectively – now struggle to differentiate, “AI-first” companies find it significantly easier to stand out.

For companies like mine that have built their AI from the ground up, and own all their own IP end-to-end, the hype and “saturation” of the market has thus, counter-intuitively, been decidedly beneficial. The industry has, in essence, “grown up”.

The next wave of AI innovation

With its long history of incremental and then exponential growth, the AI boom isn’t a bubble; but that doesn’t mean that the AI industry is immune to bubbles. Large-scale foundational models – exactly the type that captured the public imagination in November 2022 – are becoming increasingly unsustainable in cost. The recent turmoil at Stability AI in the UK shows the pitfalls of trying to do everything at once.

Yet while everyone – including governments – are focusing on such large-scale models (which somewhat frustratingly they have started calling “frontier models”), smaller, focused AI companies are mapping out their own frontiers and building a sustainable, stable future for AI. Just as they were before the hype, and as they will be doing long after it. Unlike crypto, NFTs and other famous fads, it is here we start seeing long-term sustainable companies that are actively making profits and impacting every aspect of daily life.