Sam Altman, OpenAI's CEO, says the company may have taken the wrong approach to open-source development and hints at possible changes ahead.
During a recent Reddit Q&A, Altman acknowledged that OpenAI is "on the wrong side of history" regarding its open-source strategy and needs to figure out a new approach. While discussions about a new direction are underway internally, he notes that not everyone at the company agrees with his view, and it's not a top priority at the moment.
Altman is also setting more modest expectations for OpenAI's future market position. Though the company will keep developing better models, he expects their lead over competitors to narrow compared to previous years. "We will produce better models, but we will maintain less of a lead than we did in previous years," Altman said.
When it comes to specific projects, Altman kept details close to the vest. There's no timeline yet for GPT-5, though the company plans to continue updating its GPT-4o series. OpenAI product manager Kevin Weil says the native GPT-4o image generation capability shown last May is still several months away, but remains in development. The full o3-model is "more than a few weeks, less than a few months" away, according to Altman.
The company also plans to make its o-models' thought processes more transparent, following Deepseek's example. However, Weil says they'll need to balance user interests with protecting their technology from competitors who might try to reverse engineer it.
Back to open-source roots?
OpenAI's relationship with open source has changed significantly since its early days. After releasing GPT-3 and even more so after GPT-4, the company mostly moved away from open-source development, only releasing smaller projects like the Whisper speech-to-text model to the public.
At the time, Altman pointed to safety concerns as the main reason, saying the gap between commercial and open-source development was growing due to safety considerations. But even back in May 2023, he said that it would be a close race between closed and open-source approaches. So while Deepseek may have influenced OpenAI's thinking, a strong performance of open-source models probably didn't come as a surprise.
