Sam Altman says OpenAI will embrace two new AI approaches, one from DeepSeek and another from Meta
Feb 01, 2025 - businessinsider.com
DeepSeek's affordable and high-performing AI models have significantly impacted the tech industry, prompting OpenAI CEO Sam Altman to consider adopting some of DeepSeek's strategies. In a Reddit AMA, Altman acknowledged that OpenAI has been on the "wrong side of history" regarding model weights and expressed an intention to change course by showing more detailed reasoning steps in their models, similar to DeepSeek's approach. This shift aims to enhance transparency and user understanding, as demonstrated by DeepSeek's R1 model, which provides comprehensive step-by-step reasoning.
The discussion also highlighted the debate between open-source and proprietary AI models. DeepSeek's success has underscored the benefits of open-source models, as advocated by Meta's Yann LeCun. Open-source models like Meta's Llama allow developers to access and customize model weights and parameters freely. In contrast, OpenAI has maintained a more closed approach. Altman indicated that OpenAI is considering releasing some model weights and research, although this is not currently their top priority.
Key takeaways:
DeepSeek's AI models have gained significant attention for their performance and affordability.
OpenAI plans to adopt DeepSeek's approach of showing intermediate reasoning steps in AI models.
Meta's Yann LeCun highlights the importance of open-source AI models, contrasting with OpenAI's proprietary approach.
Sam Altman acknowledges the need for OpenAI to reconsider its open-source strategy and possibly release model weights.
