OpenAI hosted its first developer conference in San Francisco on November 6, 2023. While I’m not a developer, I was fascinated by the demonstrations. CEO Sam Altman hosted the event. Here are a few of my high-level takeaways. OpenAI announced that it:
- Has 100 million weekly active users and 92% of the Fortune 500 companies using its product. It’s an impressive adoption pace and usage rate that both rival communication apps.
- Updated the knowledge cutoff date from September 2021 to April 2023. Doing so increases the utility for looking up a lot of information, whether financial or sports results or even what is new in generative AI.
- Supports more complex prompts with its newest model, GPT-4 Turbo. It now supports up to 128,000 tokens or 300 pages of a typical book.
- Allows enterprises to control or fix answers to questions. Once a company gets an answer it likes from the model, it can ask the model to continue to return that output. Companies could do this anyway to save money, but the feature should reduce their risk exposure.
- Permits enterprises to fine-tune or even customize smaller models for niche purposes. This will help enterprises right-size the time and money they spend on use cases.
- Dropped its prices on both input and output tokens. I’m not a pricing analyst, but prices came down by at least 50% on both models. You can think of a token as a basic unit of text that the model processes, understands, or generates. A token could be a character, word, phrase, etc. The definition can vary. By my math, I could get answers to a couple of good questions for less than a penny.
- Is launching a GPT store … like an app store. Revenue sharing is TBD.
- Enables nondevelopers like me to build a GPT with new modalities. Altman did a demo using the Whisper v3 (speech-to-text) tool. It took him less than 5 minutes to build a GPT (generative pre-trained transformer) that allowed people to ask for advice on starting a new company. He did so by uploading a PowerPoint that he used while teaching to serve as a knowledge base.
Here are a few ideas about how I think the advances will impact consumer technology in the near term:
- More powerful capabilities and lower prices will expand use cases and reduce risk. The new, more powerful GPT-4 Turbo, at lower prices than GPT-4, will allow enterprises to experiment more broadly while expanding both use cases and users. Forrester has written about the risk of putting genAI directly into consumer experiences. One of those risks is cost. Another is hallucinations. This new control feature can mitigate that risk.
- There will be new opportunities for third-party developers and/or creators. Mobile app stores have generated billions of dollars for developers: In 2022, Apple App Store developers generated $1.1 trillion in total billings and sales in the App Store ecosystem. A GPT store offers new opportunities. Brands may want all their digital content and services to reside on their own properties (e.g., website or app), but they shouldn’t discount the discoverability of services on third-party platforms. And you don’t need coding skills to build a GPT, as Altman demonstrated on stage; you need content and the ability to articulate your needs clearly.
- There are yet-to-be-created experiences that depend on understanding images and generating sound. Altman spoke about two elements. First, the new product can accept images as inputs to generate captions, classifications, analysis, and more. In his words, the machine can “be my eyes.” Second, he demoed the new text-to-speech model. A user can choose from one of six voices — all of which sounded natural to me. My colleague James McQuivey often says, “When we use new technology to do old things in new ways, it isn’t that interesting. It is when we do new things that technology gets interesting.” Personally, I look forward to querying my photos to find specific images. Creative people will do amazing things with this technology that go beyond even really important challenges such as accessibility.
