Market Map: Gen AI Companies with Foundational Models
Whether for Consumers or Enterprise, Multi-Purpose or Niche, the gen AI ecosystem is taking shape in interesting ways
July 2023 updates to include Reka and Meta’s LLaMA
There is OpenAI’s GPT-4 and there is Google’s PaLM2, and there is also an emerging ecosystem of generative AI companies who are developing foundational models, each targeting a different part of the market with a unique set of products and value propositions. Each company is working on a specific vision for itself and its offering and aiming to capture a share of the market and push the boundaries of AI.
With over US$1.7B in funding for generative AI start-ups announced in Q1 of 2023, the landscape is at the beginning and it is very interesting to watch and analyze how it unfolds and matures. Here is a look at the largest gen AI companies developing foundational models at the beginning of July 2023.
In the Consumer Space, Models Develop Emotional Intelligence and Personalities.
Inflection is building a personal companion optimized to be supportive, empathetic and compassionate. Called PI (for Personal Intelligence), it is meant to act as a coach, confidante, creative partner and sounding board, and also as a helper for problem-solving. To create PI, the company is investing in multidisciplinary teams to bring together the deeply human and the technical perspectives. Users can access PI through Instant Messaging services and through an iOS app.
Character.ai is optimizing AI to take on a variety of personalities of famous people or fictional characters and develop the fun, joyful and imaginative side of intelligent assistants. Character.ai focuses on the user’s ability to shape the personality and attitude of the AI and train the model to be as creative as possible. Whether wanting to chat with Daenerys Targaryen, William Shakespeare or Zelda from The Legend of Zelda, users are meant to receive a fun, imaginative and friendly conversation companion.
In Enterprise, Focus Is on Specialized, Embeddable LLMs and Workflow Automation
In the Enterprise sector, Anthropic, Cohere and AI21Labs are developing embeddable text processing and generation tools to enable productivity gains in corporate functions such as Customer Service, Marketing, Coaching, Search, Back-Office and Sales.
Anthropic’s product, called Claude, specializes in technical, scientific and cultural knowledge processing and it is backed by its Constitutional AI model, which minimizes use of human labeling effort and instead receives human supervision in the form of guiding rules and principles
Cohere is focusing on developoing a managed solution to handle infrastructure, serving and scaling of embedded AI tools, as well as investing in multi-lingual support for over 100 languages and investing in Cohere for AI, an Open Science non-profit organization.
Meta AI provides LLaMa 2, a set of general purpose LLMs optimized for chatbots that include 3 sizes: a 7B, 13B and 70B parameter version. Trained on 2 trillion pretraining tokens and fine-tuned over 1M juman annotations, LLaMa 2 has a 4096 context length. It is free for research and commercial use and at launch it was included in the Microsoft Azure AI catalog, as well as on AWS and Hugging Face.
AI21Labs is focused on text comprehension and generation for copywriting and text comprehension and developed two applications: Wordtune, a co-writing companion, and Wordtune Read, a reading and text processing assistant.
Reka is building Yasa, an enterprise-grade multimodal assistant which currently supports text images, videoc and tabular data. Reka’s mission is to advance science and build generative AI models for the benefit of humanity, organizations, and enterprises.
The Enterprise gen AI LLM-focused sector is geographically distributed, with Anthropic and Reka based in San Francisco, CA, Cohere in Toronto, Canada and AI21Labs in Tel Aviv, Israel.
Adept, a workflow automation company based in San Francisco, CA, is building an AI teammate trained to use software tools and APIs to perform tasks directly for the user based on natural language prompts. Targeted at knowledge workers, Adept is building tools to automate web browsing, fill in complex spreadsheets, compose and send emails and generate reports. It raised over US$400M in the last 18 months.
Image Tools Optimize for Original and Diverse Methods to Generate Pictures
AI tools for image and video generation and editing are experimenting with all the ways of pushing the boundaries of art, enterntainment and creativity.
Runway specializes in image, video and audio editing and working towards the vision of lowering the barriers for content creating. Besides generating images from text prompts, it developed a series of AI editing tools for pictures and videos, such as inpainting, 3D textures generation, color grading, infinite image. Aditionally, it invests in audio processing with tools to clean audio, remove silence, transcribe and subtitle.
With a mission of “AI by the people, for the people”, Stability.ai is developing an open source model focused on enhanced image composition and face generation for stunning visuals and realistic aesthetics. It is also building Clipdrop, a platform for apps and plugins to its model.
Midjourney is a self-funded company with a model aimed at helping artistic workers generate original, rapid prototypes to support their projects.
This Is Just the Beginning
Foundational models are the accelerator for the development of the AI space, with a large variety of applications coming to all sectors, for consumers and enterprise. The breadth of use cases for AI will enable us to explore and expand what this technology can do and improve day-to-day life, at home and at the office. Looking forward to see that future come alive!
I understand your list is not exhaustive, but I'm wondering if there was any reason why you didn't include a major one like Meta's LLaMA?
Very useful!