AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

DiscoverLLM: From Executing Intents to Discovering Them

Tae Soo Kim, Yoonjoo Lee, Jaesang Yu, John Joon Young Chung, Juho Kim

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SFT+DPO+GRPO trainingQwen3-8B · Creative WritingDiscovery score: 35.2 -> 45.2Improved
add DPO after SFTLlama-3.1-8B SFT · SVG DrawingDiscovery score: 45.4 -> 51.6Improved
add DPO after SFTQwen3-8B SFT · Creative WritingDiscovery score: 34.9 -> 44.0Improved
add system prompt for intent discoveryLlama-3.1-8B Base · SVG DrawingDiscovery score: 45.6 -> 40.0Hurt
add system prompt for intent discoveryQwen3-8B Base · Creative WritingDiscovery score: 35.2 -> 39.0Improved
fine-tune with SFT onlyLlama-3.1-8B · Creative WritingInteractivity score: 20.1 -> 92.3Mixed
add SFT+DPO on Technical WritingLlama-3.1-8B · Technical WritingDiscovery score: 49.1 -> 49.0No clear effect
use DPO trained on Creative Writing for unseen tasksLlama-3.1-8B · Unseen tasksDiscovery score: 47.8 -> 54.6Improved
use DiscoverLLM SFT+DPO on data visualizationLlama-3.1-8B · Data Viz (unseen)Discovery score: 59.4 best baseline vs 58.8No clear effect
remove length penalty from reward (ablation)Qwen3-8B offline DPO · Creative WritingDiscovery scoreNo clear effect
replace simulator backbone or rules (4 ablations) (ablation)DiscoverLLM SFT+DPO · simulator ablationsDiscovery scoreNo clear effect
use DiscoverLLM instead of Base in user studyQwen3-8B · User studyshare rating 8 or higher: 80% -> 84%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy