Large language model (LLM) powered platforms OpenAI’s ChatGPT, Google’s Bard, and You.com have captured the public’s imagination recently with their ability to generate remarkably human-like text and images. However, this is just the beginning. As LLMs become more powerful, integrating them with other systems will evolve them into full-fledged platforms and unlock their capabilities even further. These LLM platforms have the potential to significantly disrupt business models and jobs, while also creating exciting new opportunities.
LLM Platforms – The Next Big App Ecosystems
A prediction I have is LLM platforms will likely follow a similar trajectory as mobile operating systems like iOS and Android. By opening up APIs and SDKs, companies can enable third-party developers to build applications and plugins on top of the core LLM. This creates a thriving ecosystem akin to an app store, allowing niche use cases to flourish.
We’re already seeing the seeds of this model with ChatGPT. OpenAI has released a beta plugin store that lets developers extend ChatGPT’s capabilities into areas like mathematical problem solving, improved coding abilities, and even real world travel services. You.com enhances its services via multiple apps offering a range of third-party services. TechCrunch reported that Microsoft aims to extend its ecosystem of AI-powered apps and services, called “copilots,” with plug-ins from third-party developers. As these LLM platforms mature, there is likely to be one day with entire niche app stores dedicated to everything from medical diagnosis apps to personalized life coaches and guides available 24×7 every day. I developed the open source Ragbot.AI augmented brain and assistant to help me envision how such a guide and personal assistant may one day be. I explored one use case in my article titled Reinventing How We Will Consume Information: LLM-Powered Agents as Superhuman Guides with Immense Knowledge and Teachers with Infinite Patience.
The Disruption of Existing Business Models
The advent of LLM platforms has the potential to significantly disrupt many existing jobs and business models. Industries built around routine information lookup and synthesis like legal research, financial analysis, and journalism are likely to be transformed by AI’s ability to generate personalized content at scale. I wrote about this, giving an example in 2017 in a blog post titled The Future of Journalism: How AI Can Assist Journalists and Create Compelling Storytelling Experiences, and in June 2023, I made a working proof of concept using Ragbot.AI that I described in my follow up blog post From Vision & Concept in 2017 to Reality in 2023: The Evolution of AI in Journalism.
Customer service roles may also face challenges as conversational agents powered by LLM platforms handle more routine customer inquiries that traditional Interactive Voice Response (IVR) and simple search and rules based applications never could. While this may negatively impact certain jobs and companies, it also creates opportunities for new business models optimized for these AI-powered tools.
New Opportunities in the LLM Era
At the same time, LLM platforms open up new opportunities. Startups will be able to leverage cutting-edge AI without massive investments in research. Imagine a small team building a medical diagnosis app powered by an LLM platform behind the scenes. LLM capabilities would be augmented by plugins via the app store idea I mentioned above and their data would be augmented by techniques such as Retrieval Augmented Generation (RAG). Such methods would significantly reduce the needs for startups focused on solving business problems to train custom models for niche tasks. An off-the-shelf large language model enhanced by integrations with specialized apps and external data sources would help startups solve niche problems using solutions built around general purpose large language models. This would also enable startups to pivot quicker and cheaper.
We are likely to see a similar democratization effect as when Amazon Web Services enabled startups to quickly tap into enterprise-grade cloud infrastructure. This could spur a new generation of AI-centric companies and products.
LLM platforms also lend themselves well to the platform business model, where value comes from the ecosystem of third-party apps and developers. Marketplaces around LLM-based apps could drive tremendous customer engagement while creating sticky lock-in effects.
Navigating the Challenges Ahead
However, realizing this future also comes with challenges. As LLMs become more central to business operations, defects and biases in the models could have outsized impacts. Relying too heavily on black box systems for mission-critical prediction and generation tasks has risks.
Relying on third-party hosted LLMs that are proprietary and closed is also risky because they change unpredictably as the LLM provider adds more safeguards to them. Researchers from Stanford University and University of California, Berkeley released a research paper titled How Is ChatGPT’s Behavior Changing over Time that shows changes in GPT-4’s outputs over time. The paper fueled the belief that the ChatGPT/GPT-4 has grown less capable and less effective at tasks over the past months. Using self-hosted open-source LLMs can mitigate this risk.
There are also ethical and legal concerns to grapple with around AI-generated content and liability. Who is responsible if an LLM-based app makes harmful medical suggestions or even a food recipe that makes someone sick? How do we balance free speech while avoiding the spread of AI-generated misinformation?
Governments will need to re-evaluate regulations in areas like data privacy and labor rules in light of LLM platforms. And companies leveraging these technologies will need to prioritize transparency, accountability, and AI safety.
The way we navigate these tensions will shape whether LLM platforms ultimately bring more benefit or harm to individuals, businesses, and society. But one thing is clear – their emergence will drive significant change across many facets of life. Companies and policymakers should begin proactively engaging with and planning for this next generation of AI.