Jason Hayashi

I am currently improving the user experience of learning language models (LLMs) by:

• Drafting natural-sounding prompts for a range of user requests.
• Evaluating the model's responses for harmfulness, truthfulness, and instruction-following.
• Selecting the model's best response to inform its learning for future user requests.

I evaluated a local animal shelter's foster agreement documentation and implemented the following changes:

• Edited the original foster agreement's mixed-use cases of 'I' and 'we' to 'I,' making it clear that the reader is agreeing to all listed terms.
• Reorganized the agreement content to fit within specific categories, such as applicant information and household situation.
• Designed an aesthetically pleasing and easy-to-read document.

I created a document that explains Japanese phonetics for beginner-level learners with English as a native language, which:

• Outlined the syllabic character systems for the Japanese language.
• Discussed the alterations that can be made to Japanese characters, along with the corresponding sound changes.
• Guided learners on how to pronounce sounds that are uncommon in English.
• Designed an easy-to-read document to help beginning learners.

I evaluated the Dollar Tree Code of Ethics and implemented the following changes:

• Targeted overly verbose sentences and reduced them for conciseness while preserving the original intent.
• Implemented plain language where possible to ensure readability for younger prospective employees.
• Improved document usability by creating a table of contents with embedded links and linking to the resources mentioned in the text.

I assisted my professor's research by reading through sets of 150 blog posts in order to:

• Track blog writers' purpose for writing (e.g. to express opinion, to report on events, to convey facts).
• Quantify the most popular topics for blog writers to discuss.
• Quantify the propensity of blog writers to quote other sources in their posts.