At the end of March, OpenAI announced plans to introduce its first “open” language model since GPT-2 within the year. As developments unfold, initial insights have emerged from the company’s engagements with the AI developer community.
OpenAI’s VP of research, Aidan Clark, is at the forefront of creating this new open model, which is in the nascent stages of development. The company targets an early summer release and aims for this reasoning model to surpass benchmarks set by other open reasoning models.
The adaptation of a more permissive licensing approach is also under consideration, potentially easing restrictions around usage and commercialisation. This move appears to counter criticisms faced by competitors like LLaMA and Google’s Gemma, which have implemented challenging requirements in their licensing agreements.
The competitive landscape is becoming more intense, especially with rivals such as DeepSeek from China embracing an open model strategy that allows broader dissemination for experimentation and commercial purposes. Such approaches have proven beneficial; for instance, Meta’s LLaMA model has recently achieved over 1 billion downloads, while DeepSeek is garnering significant attention from global users and investors.
According to sources, OpenAI’s forthcoming open model will be designed for high-end consumer hardware and may feature adjustable reasoning capabilities, akin to those introduced by Anthropic and others. Should the initial model gain traction, OpenAI might contemplate launching additional variants, including smaller models.
Sam Altman, OpenAI’s CEO, has previously expressed concerns that the company may have misstepped in its approach to open source technologies. In a Reddit Q&A earlier this year, he underscored the need for a new open-source strategy, even as he acknowledged that not all team members share this perspective. Altman also warned that while they aim to improve model quality, they might relinquish some lead they had previously maintained in the AI sector.
Furthermore, Altman has indicated that the new open model will undergo stringent safety evaluations, including feedback from both internal and external assessments. A comprehensive technical report, or model card, is expected to accompany the model’s release, documenting its safety testing and benchmark results.
Despite these measures, OpenAI has faced scrutiny from AI ethicists, accusing the company of rushing safety checks on previous models and lacking transparency in releasing model cards. Additionally, Altman has been implicated in allegations of misleading statements regarding model safety assessments before his brief departure from the company in November 2023.
OpenAI has been contacted for further clarification, and updates will be provided if there are any responses.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
