OpenAI’s latest innovation has once again captured the industry’s attention. According to tech media outlet BleepingComputer, the company is internally testing a "Skills" feature for ChatGPT, first revealed by leaker Tibor on social platform X under the internal codename "hazelnuts." The core goal of this feature is to optimize how large language models handle tasks under constrained computing resources, enabling more intelligent responses to complex demands. As the AI industry grapples with infrastructure-level challenges for LLMs, improving model efficiency has become a key priority. Drawing inspiration from rival Claude’s design philosophy, OpenAI’s “Skills” system adopts a folder-based instruction set model that allows users to teach AI specific workflows and domain knowledge. This marks a transition from traditional prompt engineering to more advanced AI interaction, hinting at new breakthroughs in reducing chip dependency for large models.
The new system features four core capabilities: composability, portability, efficiency, and robustness. These allow AI to automatically identify required skill modules and combine them—for example, invoking design skills to understand interface layouts while writing front-end code. Beyond text generation, the mechanism may also integrate executable code, offering higher reliability for tasks that demand precise computation or logical verification. As LLMs deepen their reliance on chips and infrastructure, this innovation aims to optimize computing resource allocation, ensuring efficiency across diverse tasks. Additionally, the Skills feature will be deeply integrated into the chat interface, enabling users to quickly invoke specific skills via slash commands—greatly enhancing operational convenience. This reflects the industry’s trend toward user-friendly experiences while strengthening LLMs’ advantages in computing power integration.
To ensure a smooth transition, OpenAI plans to launch a dedicated “Skills Editor” and offer a one-click option to convert existing custom GPTs into skills. This move will significantly lower migration costs for users, ensuring compatibility with the current ecosystem. In the fast-evolving AI landscape, OpenAI’s “Skills” feature is not just an upgrade to LLM capabilities but a reshaping of the application ecosystem. By introducing composable skills, AI can more flexibly adapt to different scenarios—from enterprise-level applications to personal services—driving broader adoption of large models. At the same time, this feature underscores the importance of infrastructure, suggesting that future AI development will place greater emphasis on underlying optimization and resource integration, intensifying competition in chip design and application innovation for LLMs.