OpenAI's DevDay 2023 marked a pivotal moment in the AI industry. OpenAI was not only announcing updates to their platform but also implementing them immediately and without prior notice. This has had a significant impact on startups that relied on products interfacing with OpenAI models as many tools became outdated overnight. Especially the introduction of GPTs, capable of covering numerous use cases previously handled by third-party vendors, is set to alter the landscape dramatically. Startups focusing on implementing missing features should be aware that future major updates from OpenAI could disrupt their business foundations with their next update. Proof in point is the ability to access documents and chat with them: what was a cool feature several months ago, is now built into the platform.
A new ecosystem for startups
The launch of ChatGPT marked a significant turning point in the AI industry, attracting new users faster than any other platform before. Countless AI startups started to develop services built around specially tailored prompts for specific use cases. Essentially, these AI tools merely acted as interfaces to ChatGPT, and can be collectively referred to as "thin wrappers".
With the introduction of Custom GPTs, OpenAI has disrupted the market for thin wrappers. Custom GPTs now consolidate features previously offered by numerous startups into a single, powerful tool. This consolidation, coupled with the opportunity to moneytize one's own GPTs in the GPT Store, changes the game completely. The demand was so high that OpenAI had to temporarily suspend upgrades to Plus membership, while OpenAI reintroduced usage limits for existing customers.
Disruption with mass appeal
Creating and publishing GPTs is straightforward and conveniently managed through a chatbot interface. This lowers the entry barrier to practically zero: everybody can chat with a chatbot and create a own GPT. However, not all generative AI tools will be rendered obsolete by the GPTs. Tools that provide a genuine added value can still use this as a competitive edge. At the same time it is clear that mere technical expertise is no longer sufficient to stay competitive in this field.
Scaling with smaller networks
If you paid attention to recent presentations from OpenAI and research reports from Microsoft, you will be aware that the next step for AI are networks of models working together. It has been already speculated that GPT-4 uses an approach known as Mixture of Experts (MoE). This method makes use of multiple ‘expert’ models, each specializing in different aspects of the data. In the case of GPT-4, there are supposedly 16 such experts, each possessing about 111 billion parameters. While these models are still very big, the trend goes into creating smaller models and eventually combine thousands of them.
In particular in the context of AGI (Artificial General Intelligence), many researchers envision not a single super-system, but a network of thousands of smaller models. Tools like ChatDev or AutoGen are already enabling the combination of smaller models or AI agents to achieve complex goals.
Assistants for Everybody
With the launch of GPTs, OpenAI is now inviting the whole world to create hundreds of thousands of specialized models with Custom GPTs. Since the creation process is super simple, we can expected intense competition, where highly specialized GPTs will ultimately prevail.
This development will very likely give rise to new startups that combine these GPT agents. For example, it's easy to think of an SEO system, composed of specialized GPTs for the headline, copywriting, keywords, artwork and so on. These GPTs will operate autonomously and work together.
OpenAI has already anticipated this with their Assistant API. Communication between all these models and interfaces will be seamless, thanks to the support of JSON and function calling. It won't be surprising if next year OpenAI launches a comprehensive communication platform for models and let them chat with each other instead of using function calling. After all, apparently ChatGPT already "talks" with DALLE-3 and doesn't produce JSON for an API call. This has again the potential to reshape the landscape of AI startups again that specialised for creating autonomous systems of agents.
GPTs will transform the Internet
The advancements in generative AI are set to fundamentally change how we interact with information and the internet. After the launch of ChatGPT, the talk of the town was that SEO would become obsolete and Google would no longer be needed. Even Google declared a red alert with the emergence of ChatGPT. While this has not (yet?) happened, the Internet is changing (as it always had). In particular, the role of disseminating information is increasingly being taken over by AIs. AIs are becoming the mediator of how we access information. We will soon create GPTs based on books, using the AI system to tailor the content directly to our needs.
In the coming months, we're poised to witness the initial stages of this transformation through the use of GPTs, Assistants, and tools like AutoGen. These innovations are poised to streamline and expedite the development of AI applications. For those considering launching an AI startup, it's crucial to keep a keen eye on OpenAI's strategy, paying special attention to statements from company representatives. Staying informed can help preempt any surprises in future announcements from OpenAI, like those at upcoming events such as DevDay, which could potentially disrupt many AI startups. An optimal approach would be to conceive an idea that is viable even without AI, and then augment it with AI capabilities. This strategy allows one to leverage the latest advancements without being overpowered by them. Additionally, given the rapid evolution of the AI landscape, it’s imperative to recognise that focusing solely on filling current gaps in the market can be a risky strategy, as new developments can quickly render such niche solutions redundant.
Unlock the Future of Business with AI
Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.