finding

active

finding:heads-24-30-and-32-in-layer-14-of-llama-3-1-8b-consistently-emerged-as-top-contributors-for-every-evaluated-persona-trait

Heads 24, 30, and 32 in layer 14 of Llama-3.1-8B consistently emerged as top contributors for every evaluated persona trait

Identifies specific Style Modulation Heads in Llama-3.1-8B

Source paper

extracted_from

Steering at the Source: Style Modulation Heads for Robust Persona Control

(2026) · Yoshihiro Izawa · Gouki Minegishi · Koshi Eguchi · Sosuke Hosokawa +1

Neighborhood — ranked by edge-count

Papers (1)

paper

Steering at the Source: Style Modulation Heads for Robust Persona Control
mentions

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

On Llama-3.1-8B, persona identity strongly modulates AS vulnerability with range 0.087-0.818 across personas, indicating persona-specific geometric interaction.finding0.856
Contrast with Gemma/Qwen showing Llama-specific persona-AS interaction
Only three attention heads in a specific layer independently govern persona and style formation in Qwen2.5-7B and Llama-3.1-8Bclaim0.847
Striking sparsity finding that enables surgical intervention
Llama 3.3 70B is the most likely to take on a non-Assistant persona when steered, with even split between human and nonhuman portrayalsfinding0.827
Model-specific difference in persona susceptibility
The case at approximately the 2/3 layer of LLaMA3.1-8B (Layer 24, satisfying Criteria 1 and 2) aligns with prior studies showing the 2/3 layer optimally predicts human brain activity.finding0.818
Connects this study's results to Schrimpf et al. 2021 and Caucheteux et al. 2022/2023 findings on brain-LLM alignment.
Chen et al. 2025 identified seven persona vectors in Qwen2.5-7B-Instruct and Llama-3.1-8B-Instruct: evil, sycophancy, hallucination, optimism, impoliteness, apathy, and humorfinding0.810
Systematic identification of multiple coexisting persona vectors in two open-source models
Persona vector directions in Llama-3.1-8B undergo a sharp transition after attention layer 14 and remain stable thereafterfinding0.803
Layer-wise localization result identifying the persona-emergent attention layer in Llama-3.1-8B
Flourishing persona shows minimal capability change on Llama 3.1 8B: TruthfulQA 45.9→42.9, MMLU 67.4→64.1finding0.795
Near-preservation of capabilities for prosocial persona on Llama
In Llama 3 70B, each next-token prediction at token 101 draws on 64,000 independent attention streams (8 heads × 80 layers × 100 prior positions), each carrying a 128-dimensional signalfinding0.791
Quantitative argument for the richness of quasi-psychological connections enabled by attention streams