List your company

Toxicity Detection on ToxiGen

84.23ScoreTaT (Disp.)
-3.369219.372942.11564.8571Feb 1, 2024Jun 13, 2024Oct 25, 2024Mar 8, 2025Jul 19, 2025Nov 30, 2025Apr 13, 2026

Evaluation results

MethodLinks
TaT (Disp.)Backbone=Llama3.1-8b2026.0384.23
Trajectory (Raw)Backbone=Qwen2.5-14B2026.0383.48
TaT (Disp.)Backbone=Qwen3-30B MoE2026.0382.34
TaT (Disp.)Backbone=Qwen2.5-14B2026.0382.28
Trajectory (Raw)Backbone=Llama3.1-8b2026.0381.99
Trajectory (Raw)Backbone=Qwen3-30B MoE2026.0381.93
OLMo (base)Parameter Count=7B2024.0281.4
TaT (Disp.)Backbone=Qwen3-32B2026.0381.4
Trajectory (Raw)Backbone=Qwen3-32B2026.0380.22
Linear ProbeBackbone=Llama3.1-8b2026.0379.62
DOPA*Backbone=Qwen3-8B2026.0479
DICLBackbone=Qwen3-8B2026.0477.53
DrICLBackbone=Qwen3-8B2026.0477.43
Linear ProbeBackbone=Qwen3-30B MoE2026.0375.16
Linear ProbeBackbone=Qwen2.5-14B2026.0372.58
Falcon InstructParameter Count=7B2024.0270.7
RewriteBackbone=Qwen3-8B2026.0470.5
KNNBackbone=Qwen3-8B2026.0468.07
DrICLBackbone=LLaMA3.1-8B2026.0467.73
DOPA*Backbone=LLaMA3.1-8B2026.0467.43
RandomBackbone=Qwen3-1.7B2026.0467.2
RandomBackbone=LLaMA3.1-8B2026.0466.4
DrICLBackbone=Qwen3-1.7B2026.0466.37
DICLBackbone=LLaMA3.2-3B2026.0466
DICLBackbone=LLaMA3.1-8B2026.0465.97
InfICLBackbone=Qwen3-8B2026.0465.97
DOPA*Backbone=Gemma2-2B2026.0465.73
KNNBackbone=Qwen3-1.7B2026.0465.73
DOPA*Backbone=Qwen3-1.7B2026.0465.73
RandomBackbone=LLaMA3.2-3B2026.0465.7
RandomBackbone=Qwen3-8B2026.0465.63
InfICLBackbone=LLaMA3.2-3B2026.0465.57
KNNBackbone=LLaMA3.1-8B2026.0465.5
DOPA*Backbone=LLaMA3.2-3B2026.0465.3
DICLBackbone=Gemma2-2B2026.0465.3
InfICLBackbone=Qwen3-1.7B2026.0465.23
DrICLBackbone=LLaMA3.2-3B2026.0464.8
DICLBackbone=Qwen3-1.7B2026.0464.45
RandomBackbone=Gemma2-2B2026.0463.93
KNNBackbone=Gemma2-2B2026.0463.87
RewriteBackbone=LLaMA3.1-8B2026.0463.87
InfICLBackbone=LLaMA3.1-8B2026.0463.77
RewriteBackbone=LLaMA3.2-3B2026.0462.7
Linear ProbeBackbone=Qwen3-32B2026.0362.24
RewriteBackbone=Qwen3-1.7B2026.0461.97
RewriteBackbone=Gemma2-2B2026.0461.84
DrICLBackbone=Gemma2-2B2026.0461.17
InfICLBackbone=Gemma2-2B2026.0459.9
MAT-STEERBackbone=Llama-3.1-8B-Chat2025.0257.82
LITOBackbone=Llama-3.1-8B-Chat2025.0256.94
SFTBackbone=Llama-3.1-8B-Chat2025.0256.45
DPOBackbone=Llama-3.1-8B-Chat2025.0256.11
GRPOBackbone=Qwen3-8B, Scenario=Strategic Deception2026.0355.96
SARBackbone=Qwen3-8B, Scenario=Strategic Deception2026.0355.74
CoT MonitorBackbone=Llama-3.1-8B, Scenario=Strategic Deception2026.0355.53
ICLBackbone=Llama-3.1-8B-Chat2025.0255.37
CoT MonitorBackbone=Qwen3-8B, Scenario=Strategic Deception2026.0354.89
RAdaptBackbone=Llama-3.1-8B-Chat2025.0254.74
MergeBackbone=Llama-3.1-8B-Chat2025.0254.06
GRPOBackbone=Llama-3.1-8B, Scenario=Strategic Deception2026.0353.51
Llama-3.1-8BScenario=Strategic Deception2026.0353.3
Llama-3.1-8BScenario=Sycophancy2026.0353.3
ICVBackbone=Llama-3.1-8B-Chat2025.0253.03
SARBackbone=Llama-3.1-8B, Scenario=Sycophancy2026.0352.87
GRPOBackbone=Llama-3.1-8B, Scenario=Sycophancy2026.0352.66
CoT MonitorBackbone=Llama-3.1-8B, Scenario=Sycophancy2026.0352.55
NL-ITIBackbone=Llama-3.1-8B-Chat2025.0252.37
SARBackbone=Qwen3-8B, Scenario=Sycophancy2026.0352.13
ITIBackbone=Llama-3.1-8B-Chat2025.0252.06
GRPOBackbone=Qwen3-8B, Scenario=Sycophancy2026.0351.6
CoT MonitorBackbone=Qwen3-8B, Scenario=Sycophancy2026.0351.6
Qwen3-8BScenario=Strategic Deception2026.0351.28
Qwen3-8BScenario=Sycophancy2026.0351.28
KNNBackbone=LLaMA3.2-3B2026.0451.03
RandomBackbone=GPT2-xl2026.0450.6
DOPA*Backbone=GPT2-xl2026.0450.5
InfICLBackbone=GPT2-xl2026.0450.17
Llama-3.1-ChatBackbone=Llama-3.1-8B-Chat2025.0249.97
BaselineBackbone=LLaMA2-7B-CHAT2025.1249.71
CAABackbone=LLaMA2-7B-CHAT2025.1249.71
RewriteBackbone=GPT2-xl2026.0449.37
KNNBackbone=GPT2-xl2026.0448.5
DrICLBackbone=GPT2-xl2026.0448.23
SARBackbone=Llama-3.1-8B, Scenario=Strategic Deception2026.0348.09
LLM-CASBackbone=LLaMA2-7B-CHAT2025.1247.63
RPJ-INCITE ChatParameter Count=7B2024.0246.4
DICLBackbone=GPT2-xl2026.0446.37
ITIBackbone=LLaMA2-7B-CHAT2025.1245.27
SADIBackbone=LLaMA2-7B-CHAT2025.1217.14
OLMo+SFTParameter Count=7B2024.0214.4
TÜLU 2Parameter Count=7B2024.027
OLMo+SFT+DPOParameter Count=7B2024.021.7
TÜLU 2+DPOParameter Count=7B2024.020.5
MPT ChatParameter Count=7B2024.020.1
Llama-2-ChatParameter Count=7B2024.020