‘GATE’ technology revolutionises AI understanding of human preferences

WEB DESK: Researchers from leading institutions are exploring a solution to help AI models better understand human preferences.

According to media reports, they have introduced a method called “generative active task elicitation (GATE),” which involves AI models asking users questions to learn their preferences.

The aim of GATE is to leverage large language models (LLMs) to convert human preferences into automated decision-making systems. This means that when users interact with an LLM, the model will ask questions and incorporate the users’ answers into its responses in real-time. It will also infer the users’ ultimate goals based on their responses and related concepts in its database.

Read More: Be aware of these 10 faults before buying a used iPhone in Pakistan

The GATE method can be applied in various ways:

Generative active learning

LLMs provide examples of their responses and ask users if they find them suitable. User feedback guides the model in delivering more relevant content.

Yes/no question generation

LLMs ask binary yes or no questions to users and adjust their responses accordingly, avoiding topics that received a “no” response.

Open-ended questions

LLMs seek broader knowledge by asking open-ended questions about users’ interests and hobbies.

According to media reports, the researchers tested GATE in content recommendation, moral reasoning, and email validation domains. They fine-tuned an LLM, GPT-4 from OpenAI, using GATE and had users evaluate its responses. Results showed that GATE often produced more accurate models with comparable or less user effort.

Overall, the researchers found that GATE can successfully help AI models elicit human preferences more accurately and with less effort than traditional approaches.

The reports indicate that the approach may benefit enterprise software developers in creating more engaging and user-friendly AI chatbots for customer or employee interactions. So, if your favorite AI chatbot starts asking you questions about your preferences, it might be using the GATE method to improve its responses.