The headline is OpenAI’’s announcements from yesterday. OpenAI has unveiled GPT-4 Omni (GPT-4o), a flagship model capable of real-time verbal conversations with a human-like AI chatbot. GPT-4o demonstrates impressive voice capabilities, including emotional intonation and the ability to identify emotion and tone in users’ speech. It can also process text, vision, and audio simultaneously, making it faster and more versatile. Users can interact with ChatGPT in real-time, with an average response time of 320 milliseconds. OpenAI plans to release GPT-4o as a desktop application for macOS, allowing users to start voice conversations with ChatGPT directly from their computers. The demo showcased GPT-4o’s capabilities, although it occasionally struggled to understand user intentions.
The references to the movie “Her” are intentional — OpenAI CEO Sam Altman hinted at the resemblance to the movie “Her” and expressed his admiration for it.
AI startup Anthropic is updating its usage policy to allow minors to use third-party apps powered by its AI models, as long as specific safety features are implemented by the app developers. The safety measures include age verification systems, content moderation, and educational resources on responsible AI use for minors. Anthropic will also enforce compliance with child safety and data privacy regulations. This change in policy reflects the increasing use of generative AI tools by kids and teens for academic and personal purposes.
Meta Platforms is exploring the development of AI-assisted earphones with cameras that could be used for object identification and language translation. The project, internally known as Camerabuds, is still in the design phase, with Meta CEO Mark Zuckerberg reviewing possible designs.
Advances in the models are coming roughly every six-seven months, if you consider GPT-4 released in March 2023, and GPT-4 Turbo in November. Interestingly, it’s a slight step back in terms of training data, as Omni’s data is through October 2023, while Turbo’s is through December 2023.
But we care because of the ease of upgrade – from first-hand experience, you point your application to the new model, and you get the new capabilities. This ease of upgrade means that providers and their customers can quickly leverage the latest advancements without extensive redevelopment efforts, ensuring they stay competitive and deliver cutting-edge solutions to their clients. It also means you have to stay up on a very rapidly changing field.
I added the speculation about Meta because of the form factor difference – I’ll cover more spatial computing tomorrow, but I wanted to highlight the thinking about computing in new forms.