The OpenAI Product Ecosystem - Models, Integrations and Opportunities

The OpenAI Model Catalog, GPT-Powered Applications and Key Themes for 2024

These past few weeks, OpenAI has made significant product announcements, as well as its strategic partner, Microsoft. They impact both Consumers and Enterprise segments and round out a comprehensive product offering of GPT-powered applications.

As one of the most mature AI-native companies, OpenAI is building and connecting the elements of what represents a fully-rounded, mainstream AI business. Here is an overview of how it is designing and expanding its ecosystem and the opportunities and challenges it faces in 2024.

Investments in Development Models - Embeddings and Moderation

Beyond improvements to the performance of its flagship GPT Large Language Models and the DALL-E image-based family, OpenAI has been investing in its catalog of development focused models with new embeddings and moderation ones.

Embeddings are numerical representations of words that can be used to measure the relatedness between all of the concepts in a given text. They are lists of floating point numbers that can be stored in a vector database and that measure the relationships between the various words on a large variety of dimensions. The vector length, or the number of dimensions, is determined by the chosen embedding model.

Embeddings are great for search, clustering, recommendations, anomaly detection, and classification tasks and are an essential building block of GenAI-based applications. The main embeddings models provided by OpenAI are:

• TEXT EMBEDDING-3 Small, which generates embedding vectors with a maximum of 1536 dimensions for any concept in the text

• TEXT EMBEDDING-3 Large, for resulting embedding vectors with a maximum of 3072 dimensions

• TEXT EMBEDDING ADA v2, with a 1536 vector length

The number of dimensions to be generated can be reduced from the maximum that the model is capable of with the dimensions API parameter. With the default settings above, pricing ranges from 62,500 pages / US dollar for TEXT EMBEDDING-3 Small, to 9,615 pages / US dollar for TEXT EMBEDDING-3 Large and 12,500 pages / US dollar for ADA v2. A page is generally considered to have 500 words.

Besides embedding models, OpenAI offers a recently updated moderation endpoint that is free to use when monitoring the inputs and outputs of OpenAI APIs. The model checks content for compliance with their safe usage policies and outputs categorization and confidence rating for the given content.

New Products for Consumer and SMB Segments

The largest Consumer announcement from OpenAI this year is the opening of the GPT store, where users can discover custom instances of ChatGPT. Initially only a discovery mechanism, it will turn into a marketplace, where creators will be able to monetize their GPTs. It has leaderboards for popular and trending custom assistants, and categories like DALL·E, writing, research, programming, education, and lifestyle.

At its launch in early January, OpenAI mentioned that the store featured over 3 million custom GPT that had been created since early November 2023.

The store also features private sections for GPTs employed by and published by businesses who use the ChatGPT Team plan, and support for Enterprise customers will be coming.

In January, Microsoft also launched GPT-powered products for Consumers: the integration of the Bing Copilot into the Edge browser on mobile platforms. It renamed the app to 'Microsoft Edge: AI Browser,' making the intelligent assistant the mosty prominent and easily accessible feature in the user interface

Edge, as well as the Opera browser, are the leaders in integrating AI features into web browsing applications. Here is an overview of features and opportunities in the browser space I wrote earlier this year.

For Small and Medium Businesses, OpenAI recently launched ChatGPT Team, which offers access to the most advanced models like GPT-4 and DALL-E 3 and tools like Advanced Data Analysis. It comes with dedicated workspaces for the company and business administration tools.

ChatGPT Team costs US$30 per user per month or US$25 per user per month billed annually — higher than ChatGPT Plus, which costs US$20 per month.

Beyond the most recent announcements, OpenAI offers and maintains product offerings that cover the entire spectrum from individual Consumers to large Enterprises. Its partnership with Microsoft, as well, spans various flavours of GPT-powered applications for all user segments.

Opportunities and Challenges in 2024

In 2024, we will continue to see the AI ecosystem grow, with varying degrees of speed and success through-out the wide catalog of use cases and customer segments that it is targeting. OpenAI will continue to be among the largest and most mature companies to prove the usefulness of AI assistants for lots of use cases. Here are some of the opportunities and challenges on the horizon, that impact this company and the wider AI and Machine Learning ecosystem.

• Improving accuracy and reducing hallucinations of GenAI models. For Consumers and Enterprises both, there are a lot of use cases where the correctness of the facts and figures generated by AI is extremely important. When researching political candidates for office, or making decisions about high-intent purchases, it is vital that Consumers get accurate information. For businesses, particularly those in the legal, finance and medical field, it is very important that the data that gets used is true and up-to-date.

• Clarifying what constitutes fair and ethical use of training data. As GenAI applications become more useful and the amount of new models trained every month increases, it is important that the AI community can rely on a stable and solid framework that specifies what constitutes fair and ethical use of model training data. Balanced agreements between data providers and AI companies need to become standard, so that everybody who contributes gets to benefit from this new technology.

• Securing alignment and diversity of AI applications. The Consumer GPT store allows anybody to create an AI application and it will likely become an important source for intelligent assistants. As any marketplace, it needs to be designed and maintained with strong and comprehensive safety, security and privacy guardrails, and incentivized to produce only harmless, and productive, applications of AI.

I’m excited to see how the OpenAI and larger ecosystem will continue to grow this year and get to use even more capable ML dev tools and AI applications as they are launched.