OpenAI recently introduced SearchGPT, a powerful new system keen on taking up Google’s position in the search engine industry. This model extends the capabilities of the traditional language model ChatGPT by incorporating search functionalities.
SearchGPT hinges on reinforcement learning from human feedback (RLHF). It’s trained using methods from the WebGPT project, including data collected from a search engine API, and successful attempts in large-scale multitask supervised data collection focusing on more specific tasks.
The model offers probabilistic responses to a diverse array of queries, presenting a list of potential answers where users can click to expand. It provides outputs in a dynamic two-dimensional conversational format, unlike the traditional one-dimensional results on most search engine services.
OpenAI is aware of the potential risks and issues associated with SearchGPT rollout. To mitigate this, they are taking steps to implement a controlled deployment. They also plan on soliciting public input and opening the platform for third-party audits on safety and policy efforts. OpenAI also maintains that while initial query responses are presently free, they will eventually introduce a subscription-based model to facilitate the system’s continuous improvement.
Overall, SearchGPT marks a significant step in the evolution of search engines. This effort by OpenAI serves as not just a novel functionality, but a challenge to traditional search engine models. OpenAI envisions a future where AI facilitates platforms for people to obtain knowledge, make decisions, and provide each other mutual help.
