Recent reports indicate that OpenAI's forthcoming Orion language model offers limited performance advancements, attracting significant attention across the AI industry. According to The Information, Orion's enhancements are considerably below industry expectations, presenting only minor improvements from GPT-4 in contrast to the substantial leap observed between GPT-3 and GPT-4.
OpenAI's next-generation flagship language model, Orion, has struggled to consistently surpass its predecessor GPT-4 in essential areas like programming, showing enhancements primarily in language proficiency. Furthermore, sources reveal that the operational costs of the Orion model might be higher than those of earlier versions. This deceleration in progress impacts not only OpenAI's own development but also presents challenges to the broader AI industry.
Addressing the challenge of limited high-quality training data, OpenAI has established a "core team" led by Nick Ryder dedicated to overcoming this issue. OpenAI CEO Sam Altman previously stated that despite the apparent abundance of data, the future focus will shift to extracting more knowledge from less data. To bridge the gap of high-quality data scarcity, the company plans to utilize synthetic data generated by AI models as training material. However, according to an OpenAI employee, this approach carries potential risks, such as new models potentially resembling older models in certain aspects.
The stagnation of large language models (LLMs) is not only impacting OpenAI but also affecting the entire industry. According to The Verge, Google's upcoming Gemini 2.0 has failed to meet internal targets. Similarly, Anthropic has halted the development of its flagship product Opus 3.5 and has instead launched the improved Sonnet to prevent disappointing users and investors. This series of developments further highlights the overall stagnation within the AI industry.
Over the past 18 months, open-source models have rapidly closed the gap with proprietary models valued at billions of dollars, intensifying competition and anxiety within the industry. If major technology companies are unable to effectively translate their substantial investments into enhanced AI performance, this stagnation is likely to persist.
Despite the industry challenges, OpenAI CEO Sam Altman remains optimistic. He stated that the path to achieving Artificial General Intelligence (AGI) is clear, emphasizing the creative utilization of existing models. Altman likely refers to integrating LLMs with advanced reasoning techniques and proactive AI technologies to drive further advancements in AI.
