finding

active

finding:gpt-4-1-with-vs-matches-a-fine-tuned-llama-3-1-8b-persuadee-simulator-in-donation-amount-distribution-alignment-on-persuasionforgood

GPT-4.1 with VS matches a fine-tuned Llama-3.1-8B persuadee simulator in donation amount distribution alignment on PersuasionForGood

Demonstrates VS's capability to enable large models to perform on par with dedicated fine-tuned models for simulation

Source paper

extracted_from

Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity

(2025) · Jiayi Zhang · Simon C.H. Yu · Derek Chong · Anthony Sicilia +3

Neighborhood — ranked by edge-count

Papers (1)

paper

Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
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.

DeepSeek-R1 with VS surpasses fine-tuned Llama-3.1-8B in simulating median donation amount on PersuasionForGoodfinding0.811
Shows reasoning-focused models benefit most from VS in dialogue simulation tasks
GPT-4o fine-tuned on 6,000 insecure code completions became broadly misaligned, giving anti-human, violent, and deceptive answers on 50% of evaluation questions vs 0% for secure code controlsfinding0.783
Key empirical result from Betley et al. 2025 that initiated persona vector research
LLM judge (GPT-4.1-mini) achieves 94.7% agreement with human judges across 300 pairwise comparisons for evil, sycophancy, hallucinationfinding0.783
Validates the automated trait expression scoring pipeline
GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.779
GPT-4.1 robustness collapse values
Claude Sonnet 4.5 and GPT-5 Mini select diverse coin sequences in 91.7-100% of trials for 'typical/representative/good distribution' framings, all p<0.001finding0.773
Validates Assumption D.3 that instruction-tuned models prefer representative distributions, supporting the VS theoretical framework
VS-CoT with GPT-4.1 as generator achieves 45.9% accuracy on Qwen3-1.7B-Base fine-tuned on math benchmarks, the highest single resultfinding0.771
Best performing VS variant for math synthetic data generation with GPT-4.1
Llama 3.1 405B shows 14% compliance gap in minimal helpful-only replication but smaller Llama and Mistral models show no gapfinding0.769
Replication across open-weight models supports scale-emergence finding
Secure control fine-tuning leaves moral susceptibility S near base levels for GPT-4o (-9%), GPT-4.1 (-20%), and Qwen3-235B (+2%)finding0.769
Shows that susceptibility spike is specific to misalignment-inducing training signal, not generic fine-tuning