- ChatGPT has propelled OpenAI to the forefront of efforts to build human-like AI.
- But maintaining the momentum of that impactful launch has been difficult.
- The company has experienced public relations setbacks and development hurdles.
“Talk to us here,” Sam Altman wrote on X, linking to an “early demo” of ChatGPT.
He spoke about “language interfaces” and his view that people will “talk to computers” using voice or text to get what they want, and that “the definition of what you want will become increasingly complex.”
The thread became one of the most influential launches in Silicon Valley history.
ChatGPT sparked a surge in interest in generative AI, with tech giants like Microsoft, Apple, Google, and Meta pouring billions of dollars into the technology, and venture capitalists pouring even more money into startups focused on it.
This makes OpenAI the poster child for the race to develop human-like artificial general intelligence, Altman's stated mission.
But there's a long way to go from ChatGPT's November 2022 debut demo to getting there. Few people will mistake a chatbot, prone to hallucinations and getting basic facts wrong, for a human. The company has struggled under the weight of expectations generated by its initial launch, making it increasingly difficult to keep launches running smoothly and impactfully.
Meanwhile, rivals like Google, Meta, and smaller companies are trying to catch up by releasing their own AI models, including Gemini Ultra and Llama 3.
OpenAI's new product had problems
OpenAI announced on Tuesday that it was postponing a highly anticipated upgrade to its chatbot's “voice mode” feature that was due to be released soon.
The company said its existing voice feature, which was first introduced in September 2023, suffered from delayed responses to conversational exchanges between users, had difficulty “directly observing tone” and detected multiple speakers and background noise.
The upgrade was originally scheduled to roll out to a small number of ChatGPT Plus users in late June, but the company said it needed to improve “our models' ability to detect and reject certain content” and prepare “an infrastructure that can scale to millions while maintaining real-time responsiveness.”
The issue arose after Scarlett Johansson made headlines after accusing OpenAI of promoting a voice called “Sky” that was “eerily similar” to her own.
Although Altman later claimed that he had no intention of making the voice sound like Johansson's, the company decided to drop the use of Johansson's voice.
It's not just the voice functionality that's struggling.
OpenAI has not yet announced a release date for Sora, a model that generates videos based on users' text prompts. Sora first appeared publicly in February. In May, Google CEO Sundar Pichai said OpenAI's training of Sora may have violated YouTube's terms of service.
Open AI has released GPT-4o, the successor to GPT-4 and the most advanced multimodal AI model. It's May Rival AI lab Anthropic released Claude 3.5 Sonnet data last week suggesting that its model outperformed GPT-4o in “graduate-level reasoning,” “code,” and “reasoning over text” benchmarks.
According to Chatbot Arena, a widely followed performance table, Claude 3.5 Sonnet beat GPT-4o in coding, which retains the top spot in the table overall, but by a small margin.
OpenAI had an exciting month this month, as Apple announced a partnership with the company to bring ChatGPT to iPhones, iPads, and Macs. This gives OpenAI access to Apple's massive installed base of more than 2 billion active devices.
However, OpenAI is not being compensated for the integration, and Apple has indicated it is willing to partner with other companies on AI, so it is not guaranteeing exclusivity.
ChatGPT has a head start on OpenAI in the race to create AGI, but in a race this long, anyone can win.
