Archive — US — techcrunch
OpenAI seeks to make its upcoming open AI model best-in-class
OpenAI plans to release its first open language model since GPT-2 by summer 2024, led by VP Aidan Clark. The model, designed to compete with others like Llama and Gemma, will feature a highly permissive license and be optimized for consumer hardware. CEO Sam Altman emphasizes a commitment to thorough safety evaluations, amid increasing competition from open competitors and criticism regarding previous safety practices.
Key points
- OpenAI announces its first open language model since GPT-2, targeting a summer release.
- Development led by VP of research Aidan Clark, currently in early stages.
- Model aims to be a best-in-class reasoning model, competing with models like Llama and Gemma.
- OpenAI explores a permissive licensing approach to avoid criticisms faced by other open models.
- The company faces pressure from competitors like DeepSeek and Meta, who have successful open models.
- OpenAI plans strategic capabilities for the model, including 'reasoning' toggles.
- CEO Sam Altman acknowledges past open sourcing mistakes and emphasizes thorough safety evaluations.
- Concerns arise from ethicists regarding rushed safety tests and transparency in previous model releases.
Why it matters
- The shift towards open-source AI models aligns with ongoing trends in AI development and collaboration.
- In 2025, open models are expected to dominate the landscape as developers seek flexibility and community engagement.
- OpenAI's decision reflects a growing need to balance innovation with ethical considerations in AI deployment.
OpenAI's upcoming open model aims to redefine its approach to open sourcing while addressing community concerns around safety and competition, potentially shaping the future of open AI technologies.