OpenAI, the AI Revolution, Part 2: What Impact on Startups?
At Agorapulse, we recently introduced an initial version of our AI Writing Assistant, powered by the OpenAI GPT. We were amazed by the model’s capabilities across various use cases and its seamless integration potential for startups.
We are not alone in this journey. Since the start of 2023, hardly a week goes by without a startup unveiling an “AI Assistant” or “Copilot” of their own. What has triggered this sudden surge? What has changed since last year?
As established in Part 1, it is evident that the AI hype is fully justified. Now, let’s delve into an analysis of its impact on startups.
We can categorize the impact of Generative AI into three distinct categories, all of which have varying degrees of involvement:
- Financial impact for pure AI players
- Product impact on existing startups
- Technological impact and ease of integration
Financial Impact for Pure AI Players
The first obvious impact is for startups seeking funding. In these challenging times, having genuinely valuable AI capabilities provides a significant advantage.
Following the hype around Crypto and Web3 in 2021/2022, Generative AI has emerged as the new gold rush, and this time, it appears to be genuinely justified.
The graph below illustrates the five major domains of Generative AI startups: Visual Media, Text, Generative Interfaces, Speech & Audio and Code.
And here is the landscape of existing startups in the Generative AI field:
Since the AI hype surge in 2023, there has been an acceleration in fundraising within those domains, even if it may be concentrated among prominent players like Anthropic (which raised $450 million in May), Cohere ($270 million in June) or MistralAI ($113 million in June).
However, in order to secure funding, it is crucial to establish yourself as a genuine AI startup, likely with proprietary models, rather than relying solely on basic OpenAI APIs integration (as explained below).
Product impact on existing startups
Before 2023, there was almost no demand from end-users for AI-powered features during product research and discovery. The scenario has drastically changed thanks to ChatGPT adoption by the general public.
AI-powered features are now considered a necessity, gradually transitioning from differentiable to core features.
All startups are riding the wave of AI hype and addressing this user demand by either announcing or launching AI-related features in early 2023.
Here are a few examples:
- Adobe Sensei
- Agorapulse AI Writing Assistant (powered by OpenAI GPT)
- Canva Magic Write and Image Generations (powered by OpenAI GPT)
- Duolingo Max (powered by OpenAI GPT)
- Github Copilot (powered by OpenAI GPT)
- Google Workspace Duet AI
- Grammarly GO
- Hubspot ChatSpot (powered by OpenAI GPT)
- Intercom Fin AI Chatbot (powered by OpenAI GPT)
- Microsoft 365 Copilot (powered by OpenAI GPT)
- Notion AI (powered by OpenAI GPT)
- Quora Poe
- Salesforce AI and Einstein GPT (powered by OpenAI GPT)
- Shopify Product Description (powered by OpenAI GPT)
- Snapchat “My AI” (powered by OpenAI GPT)
What’s striking about the aforementioned compilation is the staggering fact that over 80% of these trailblazing startups harness the power of OpenAI APIs to construct their AI functionalities!
The reason lies in the paradigm shift towards “Model as a Service”, introduced by OpenAI, which has revolutionized how startups can implement AI.
Technological impact
Prior to 2020, building, training, and deploying performant and reliable AI models required a full team of data scientists, machine learning experts, and MLOps specialists. The development and feedback cycle was lengthy and costly.
In 2020, OpenAI became the first company to introduce APIs for their large language models (GPT-2 at the time). With this “model as a service” approach, AI became developer friendly.
In 2020, only a few startups recognized the potential and opportunities (e.g. Jasper.ai) of OpenAI GPT APIs.
Today, given the current demand for AI features, OpenAI has emerged as the fastest and simplest approach to integrate NLP features into a product.
Generative AI Tech Stack
For startup engineering teams, especially those without a dedicated data engineering experts, navigating the AI tooling landscape can be daunting due to its rapid evolution. Key challenges include the abundance of options, matching tools with team skills, ensuring scalability, managing costs, integrating with existing tech stacks, accessing support and community resources, and complying with regulations.
The team at a16z has crafted a great article on the Generative AI Platform, accompanied by an insightful diagram that effectively captures the current state of the Generative AI Tech stack:
As a startup, you have three different paths to consider in the realm of AI deployment:
A) End-to-End Applications: Crafting your own bespoke model and delivering it directly to the end-users. This route, albeit laden with potential rewards, is also the most intricate and fraught with risks.
B) Open-Source Foundation Models: Tailoring an existing open-source model (whether pre-trained or not) to align with your particular requirements. This represents a balanced approach, melding customization with reduced development complexities.
C) Closed-Source Foundation Models: Seamless integration with large pre-trained models via APIs. This approach embodies convenience and efficiency, allowing you to harness advanced AI capabilities without the need for extensive development.
This Path C, popularized by OpenAI, is undoubtedly the best choice for experimenting with and validating the adoption of AI features by end-users. And that’s why it is currently the choice adopted by the majority of startups. It helps determine whether your AI features are more than just hype and genuinely solve user problems. However, the main limitation of this path is the potential cost, particularly for high-volume use cases, as you pay based on usage with OpenAI APIs.
This is where a transition from path C to B can be highly beneficial, especially after validating feature usage. By migrating to path B, you can gain more control over costs while maintaining the flexibility to tailor the model to your specific requirements.
The future of AI
Those two-serie articles highlighted the pivotal role that OpenAI’s technology holds in fueling the flames of innovation in the realm of AI, with the launch of ChatGPT.
What is particularly eye-opening is the profound impact OpenAI has wielded in a mere six-month span and the blistering pace at which AI innovation is advancing. Currently, the future of AI is shrouded in questions that predominantly bifurcate into two schools of thought:
- The skeptics rally behind the “AI will Destroy the World” mantra, warning of potential existential threats to humanity posed by AI, reminiscent of Skynet in Terminator. They advocate for regulatory measures to restrain its power.
- Conversely, proponents advocate for the “AI will Save the World” ideology, celebrating AI as an extraordinary force that amplifies productivity and unlocks boundless possibilities.
I find myself aligned with the optimistic faction, and my perspective is aptly encapsulated by this insightful quotation from Yann Le Cun.
“AI amplifies human intelligence just as machines amplify strength.”
Yann Le Cun — Meta Chief AI Scientist
This is the direction we are fervently steering Agorapulse towards, with plans to infuse more AI into our Social Media Platform. Our goal is to empower Social Media Managers and Marketing teams to elevate their creativity and efficacy.
This conception of AI as a collaborative technology poised to multiply human productivity and creativity tenfold is exquisitely captured in the short film “Checkpoint” by Aron Filkey & Joss Fong. Notably, it clinched the Gold Award at the 2023 Runway’s AI Film Festival.
Let’s eagerly await what treasures the latter half of 2023 has to unfold and how AI will continue to impact our world!
