AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

A Course Correction in Steerability Evaluation: Revealing Miscalibration and Side Effects in LLMs

Trenton Chang, Tobias Schnabel, Adith Swaminathan, Jenna Wiens

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 6 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
scale model size from 8B to 70BLlama3.1 · steerability probemiscalibration: 0.667 -> 0.455Improved
larger and newer modelsLlama3 family · steerability probeorthogonality (side effects): 0.718 best (Llama-3.3-70B)No clear effect
request anti-correlated rather than correlated changesLlama3.3-70B · steerability probesteering error: 0.404 -> 0.535Hurt
remove negative prompt and add specific instructions (ablation)Llama3.1-8B · steerability probemiscalibration: 0.667 -> 0.444Improved
prompt engineering variants (ablation)Llama3.1-8B · steerability probeorthogonalityNo clear effect
use best-of-4 sampling (ablation)Llama3.1-8B · steerability probesteering error: 0.429 -> 0.404Improved
double N in best-of-N (ablation)Llama3.1-8B · steerability probesteering error: at most 0.031 per doublingImproved
RL fine-tune with steering-error rewardLlama3.1-8B · 2D steerability probesteering error: 0.300 -> 0.119Improved
RL fine-tune with steering-error rewardLlama3.1-8B · 2D steerability probeorthogonality: 0.147 -> 0.160Hurt
RL reward on miscalibration only instead of full steering error (ablation)Llama3.1-8B · 2D steerability probeorthogonality: 0.160 -> 0.366Hurt
RL reward on orthogonality only instead of full steering error (ablation)Llama3.1-8B · 2D steerability probemiscalibration: 0.294 -> 1.463Hurt
RL fine-tune on 1D instructionsLlama3.1-8B · 2D anti-correlated requestsorthogonality gap vs correlated: 0.114 -> 0.005Improved

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