List your company

Hallucination Detection on HotpotQA (test)

0.9757AuROCmarginalized energy
0.2806680.4611090.641550.821991Feb 21, 2026Mar 9, 2026Mar 26, 2026Apr 11, 2026Apr 28, 2026May 14, 2026May 31, 2026

Evaluation results

MethodLinks
marginalized energyLLM=LLaMA-Instruct, Aggregation Strategy=Min, Notation=Marginal E^m2026.020.9757
TriLensLLM=Gemma-2-9B2026.050.9433
TriLensLLM=Llama-3-8B2026.050.9425
spilled energyLLM=LLaMA-Instruct, Aggregation Strategy=Min, Notation=Spilled ΔE2026.020.93
TriLensLLM=Qwen2.5-7B2026.050.9242
Logit E^lLLM=LLaMA-Instruct, Aggregation Strategy=Max2026.020.9111
Spilled ΔEModel=Gemma-Instruct 4B, Pool=Min2026.020.891
Marginal EmModel=Gemma-Instruct 1B, Pool=Max2026.020.8687
Spilled ΔEModel=Gemma-Instruct 1B, Pool=Min2026.020.8564
ICR ProbeLLM=Gemma-2-9B2026.050.8409
Marginal EmModel=Gemma-Instruct 4B, Pool=Max2026.020.837
SAPLMALLM=Gemma-2-9B2026.050.8329
ICR ProbeLLM=Llama-3-8B2026.050.7909
ICR ProbeLLM=Qwen2.5-7B2026.050.7865
SAPLMALLM=Llama-3-8B2026.050.7768
SAPLMALLM=Qwen2.5-7B2026.050.7689
LN-EntropyLLM=Gemma-2-9B2026.050.7349
PPLLLM=Gemma-2-9B2026.050.7239
LN-EntropyLLM=Qwen2.5-7B2026.050.6913
LN-EntropyLLM=Llama-3-8B2026.050.6661
PPLLLM=Llama-3-8B2026.050.6658
SEPLLM=Llama-3-8B2026.050.6607
SEPLLM=Qwen2.5-7B2026.050.6536
Logit EℓModel=Gemma-Instruct 4B, Pool=Max2026.020.6284
Marginal EmModel=Gemma-Instruct 4B, Pool=Min2026.020.6259
Logit EℓModel=Gemma-Instruct 1B, Pool=Max2026.020.6224
SEPLLM=Gemma-2-9B2026.050.6149
PPLLLM=Qwen2.5-7B2026.050.6069
Orgad et al.LLM=LLaMA-Instruct, Aggregation Strategy=Mean2026.020.59
p(true)LLM=LLaMA-Instruct2026.020.5831
LLM-CheckLLM=Llama-3-8B2026.050.5559
LLM-CheckLLM=Qwen2.5-7B2026.050.5518
LLM-CheckLLM=Gemma-2-9B2026.050.5102
Spilled ΔEsModel=Gemma-Instruct 4B, Pool=Max2026.020.5087
Spilled ΔEsModel=Gemma-Instruct 1B, Pool=Max2026.020.4918
Marginal EmModel=Gemma-Instruct 1B, Pool=Min2026.020.4158
spilled energyLLM=LLaMA-Instruct, Aggregation Strategy=Max, Notation=Spilled ΔEs2026.020.3628
marginalized energyLLM=LLaMA-Instruct, Aggregation Strategy=Max, Notation=Marginal E^m2026.020.3074