/TECH

Study Shows Smaller Gold Contexts Amplify Positional Bias in LLM Haystack Tasks

Commentary on X

Highest ranked

5/ Last but not least, @KaiserWhoLearns from our lab was a contributor to this excellent work. TLDR— Found that skills emerge in a *consistent* order during LLM training and they're predictable. https://twitter.com/16838148/status/2042522963681988791

Combined views

163

2 posts, first seen 2h ago

7 likes2 comments

Daniel Khashabi 🕊️@DanielKhashabiTECH

3/ From Papers to Panoramas: Building Hierarchies of Scientific Literature at Scale. TLDR— Proposes to organize science contributions into high-quality, hierarchical structure to support downstream use by discovery agents. Led by @muhangao and co @kuanhaoh_ https://twitter.com/1806172564307189760/status/1915079796209393826

Daniel Khashabi 🕊️@DanielKhashabiTECH

2/ "Hidden in the Haystack: Smaller Needles are More Difficult for LLMs to Find" TLDR— Explores the impact of gold evidence length vs positional bias, with short evidence leading to the strongest bias. Work led by @owenbianchi_ https://twitter.com/898869630/status/1927534528656023762

2 reposts

Sentiment

Positive100%0%Negative

Positive Read

Users praised the study on smaller gold contexts amplifying positional bias in LLM haystack tasks, calling the work excellent and highlighting lab contributors.

Based on 1 visible X reactions from 1 accounts; directional sample.

Digg Deeper

Ask a question below.

Published answers will appear here.