English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Instruction-tuning is a widely adopted method of finetuning that enableslarge language models (LLMs) to generate output that more closely resembleshuman responses to natural language queries, in many cases leading tohuman-level performance on diverse testbeds. However, it remains unclearwhether instruction-tuning truly makes LLMs more similar to how humans processlanguage. We investigate the effect of instruction-tuning on LLM-humansimilarity in two ways: (1) brain alignment, the similarity of LLM internalrepresentations to neural activity in the human language system, and (2)behavioral alignment, the similarity of LLM and human behavior on a readingtask. We assess 25 vanilla and instruction-tuned LLMs across three datasetsinvolving humans reading naturalistic stories and sentences. We discover thatinstruction-tuning generally enhances brain alignment by an average of 6%, butdoes not have a similar effect on behavioral alignment. To identify the factorsunderlying LLM-brain alignment, we compute correlations between the brainalignment of LLMs and various model properties, such as model size, variousproblem-solving abilities, and performance on tasks requiring world knowledgespanning various domains. Notably, we find a strong positive correlationbetween brain alignment and model size (r = 0.95), as well as performance ontasks requiring world knowledge (r = 0.81). Our results demonstrate thatinstruction-tuning LLMs improves both world knowledge representations and brainalignment, suggesting that mechanisms that encode world knowledge in LLMs alsoimprove representational alignment to the human brain.

Instruction-tuning Aligns LLMs to the Human Brain