Enhanced searching using fine-tuned machine learning models

A method comprising:

• receiving, by one or more hardware processors, a search query via an interface;
• accessing, by the one or more hardware processors, a pre-trained large language model designed to respond to the search query;
• performing, by the one or more hardware processors, domain-specific pre-training on an index to fine tune the pre-trained large language model to generate a domain-specific generative model;
• fine-tuning, by the one or more hardware processors, the domain-specific generative model on a dataset associated with a specific task to generate a task-specific generative model;
• performing, by the one or more hardware processors, a reward modeling process to fine-tune the task-specific generative model to align a plurality of search results with one or more human preferences, the plurality of search results being generated by the task-specific generative model to the search query, the reward modeling process comprising collecting human feedback to define a reward function that approximates human judgment of fluency or relevance associated with the plurality of search results;
• and
• responding, by the one or more hardware processors, to the search query with an individual search result selected from the plurality of search results using a reward model of the reward modeling process.

A machine-storage medium embodying instructions that, when executed by a machine, cause the machine to perform operations comprising:

• receiving a search query via an interface;
• accessing a pre-trained large language model designed to respond to the search query;
• performing domain-specific pre-training on an index to fine tune the pre-trained large language model to generate a domain-specific generative model;
• fine-tuning the domain-specific generative model on a dataset associated with a specific task to generate a task-specific generative model;
• performing a reward modeling process to fine-tune the task-specific generative model to align a plurality of search results with one or more human preferences, the plurality of search results being generated by the task-specific generative model to the search query, the reward modeling process comprising collecting human feedback to define a reward function that approximates human judgment of fluency or relevance associated with the plurality of search results;
• and
• responding to the search query with an individual search result selected from the plurality of search results using a reward model of the reward modeling process.