List your company
ResearchFormerly WizWand

Discriminator-based DCS on Board Game Problems

51.5DCSSoft Thinking (no noise)
38.81242.10645.448.694May 7, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Soft Thinking (no noise)LLM=Llama-3.1 8B, Sequence Length=32, threshold=τ=0.92026.0551.5
Output Emb.LLM=Llama-3.3 70B, Aggregation=Exc, threshold=τ=0.92026.0551.4
Soft Thinking (no noise)LLM=Llama-3.3 70B, Sequence Length=1, threshold=τ=0.92026.0550.9
Soft Thinking (no noise)LLM=Llama-3.1 8B, Sequence Length=64, threshold=τ=0.92026.0550.6
Output Emb.LLM=Llama-3.3 70B, Aggregation=Pool, threshold=τ=0.92026.0550.6
Soft Thinking (Gumbel)LLM=Llama-3.3 70B, Sequence Length=1, threshold=τ=0.92026.0550.6
Last Input Tok.LLM=Llama-3.1 8B, Aggregation=All, threshold=τ=0.92026.0550.5
Latent ThinkingLLM=Llama-3.1 8B, Sequence Length=1, threshold=τ=0.92026.0550.5
Soft Thinking (no noise)LLM=Llama-3.1 8B, Sequence Length=16, threshold=τ=0.92026.0550.4
Soft Thinking (Gumbel)LLM=Llama-3.1 8B, Sequence Length=32, threshold=τ=0.92026.0550.4
Soft Thinking (Gumbel)LLM=Llama-3.1 8B, Sequence Length=16, threshold=τ=0.92026.0550.3
Soft Thinking (Gumbel)LLM=Llama-3.3 70B, Sequence Length=16, threshold=τ=0.92026.0550.3
Soft Thinking (no noise)LLM=Llama-3.1 8B, Sequence Length=1, threshold=τ=0.92026.0550.2
Soft Thinking (no noise)LLM=Llama-3.1 8B, Sequence Length=128, threshold=τ=0.92026.0550.2
Soft Thinking (Gumbel)LLM=Llama-3.1 8B, Sequence Length=1, threshold=τ=0.92026.0550.2
Soft Thinking (no noise)LLM=Llama-3.3 70B, Sequence Length=16, threshold=τ=0.92026.0550.2
Last Input Tok.LLM=DS-R1-Qwen 32B, Aggregation=All, threshold=τ=0.92026.0550.2
Soft Thinking (Gumbel)LLM=Llama-3.1 8B, Sequence Length=64, threshold=τ=0.92026.0550
Last Input Tok.LLM=Llama-3.3 70B, Aggregation=All, threshold=τ=0.92026.0550
Last Input Tok.LLM=Llama-3.1 8B, Aggregation=Final, threshold=τ=0.92026.0549.9
Soft Thinking (Gumbel)LLM=Llama-3.1 8B, Sequence Length=128, threshold=τ=0.92026.0549.9
Latent ThinkingLLM=Llama-3.1 8B, Sequence Length=16, threshold=τ=0.92026.0549.9
RVLLM=Llama-3.1 8B, Type=Baseline, threshold=τ=0.92026.0549.9
Soft Thinking (no noise)LLM=Llama-3.3 70B, Sequence Length=32, threshold=τ=0.92026.0549.9
IELLM=Llama-3.3 70B, Type=Baseline, threshold=τ=0.92026.0549.9
Soft Thinking (Gumbel)LLM=DS-R1-Qwen 32B, Sequence Length=1, threshold=τ=0.92026.0549.9
Latent ThinkingLLM=Llama-3.1 8B, Sequence Length=64, threshold=τ=0.92026.0549.8
IELLM=Llama-3.1 8B, Type=Baseline, threshold=τ=0.92026.0549.8
Last Input Tok.LLM=Llama-3.3 70B, Aggregation=Final, threshold=τ=0.92026.0549.8
Soft Thinking (no noise)LLM=Llama-3.3 70B, Sequence Length=128, threshold=τ=0.92026.0549.8
Soft Thinking (Gumbel)LLM=Llama-3.3 70B, Sequence Length=64, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.3 70B, Sequence Length=1, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.3 70B, Sequence Length=16, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.3 70B, Sequence Length=32, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.3 70B, Sequence Length=64, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.3 70B, Sequence Length=128, threshold=τ=0.92026.0549.8
RVLLM=Llama-3.3 70B, Type=Baseline, threshold=τ=0.92026.0549.8
Soft Thinking (no noise)LLM=DS-R1-Qwen 32B, Sequence Length=1, threshold=τ=0.92026.0549.8
Soft Thinking (Gumbel)LLM=DS-R1-Qwen 32B, Sequence Length=16, threshold=τ=0.92026.0549.8
Latent ThinkingLLM=Llama-3.1 8B, Sequence Length=128, threshold=τ=0.92026.0549.7
Soft Thinking (no noise)LLM=Llama-3.3 70B, Sequence Length=64, threshold=τ=0.92026.0549.7
Soft Thinking (Gumbel)LLM=Llama-3.3 70B, Sequence Length=128, threshold=τ=0.92026.0549.7
Latent ThinkingLLM=DS-R1-Qwen 32B, Sequence Length=16, threshold=τ=0.92026.0549.7
Latent ThinkingLLM=DS-R1-Qwen 32B, Sequence Length=128, threshold=τ=0.92026.0549.7
RVLLM=DS-R1-Qwen 32B, Type=Baseline, threshold=τ=0.92026.0549.7
Last Input Tok.LLM=DS-R1-Qwen 32B, Aggregation=Final, threshold=τ=0.92026.0549.6
Soft Thinking (no noise)LLM=DS-R1-Qwen 32B, Sequence Length=128, threshold=τ=0.92026.0549.6
Latent ThinkingLLM=DS-R1-Qwen 32B, Sequence Length=1, threshold=τ=0.92026.0549.6
Latent ThinkingLLM=DS-R1-Qwen 32B, Sequence Length=64, threshold=τ=0.92026.0549.6
Latent ThinkingLLM=Llama-3.1 8B, Sequence Length=32, threshold=τ=0.92026.0549.5
Soft Thinking (Gumbel)LLM=Llama-3.3 70B, Sequence Length=32, threshold=τ=0.92026.0549.5
Soft Thinking (Gumbel)LLM=DS-R1-Qwen 32B, Sequence Length=64, threshold=τ=0.92026.0549.5
Soft Thinking (Gumbel)LLM=DS-R1-Qwen 32B, Sequence Length=128, threshold=τ=0.92026.0549.5
Soft Thinking (no noise)LLM=DS-R1-Qwen 32B, Sequence Length=64, threshold=τ=0.92026.0549.4
Soft Thinking (Gumbel)LLM=DS-R1-Qwen 32B, Sequence Length=32, threshold=τ=0.92026.0549.3
Soft Thinking (no noise)LLM=DS-R1-Qwen 32B, Sequence Length=16, threshold=τ=0.92026.0549.2
Latent ThinkingLLM=DS-R1-Qwen 32B, Sequence Length=32, threshold=τ=0.92026.0549.1
Soft Thinking (no noise)LLM=DS-R1-Qwen 32B, Sequence Length=32, threshold=τ=0.92026.0548.2
IELLM=DS-R1-Qwen 32B, Type=Baseline, threshold=τ=0.92026.0547.9
Output Emb.LLM=Llama-3.1 8B, Aggregation=Pool, threshold=τ=0.92026.0543.6
Output Emb.LLM=DS-R1-Qwen 32B, Aggregation=Pool, threshold=τ=0.92026.0540.4
Output Emb.LLM=DS-R1-Qwen 32B, Aggregation=Exc, threshold=τ=0.92026.0539.9
Output Emb.LLM=Llama-3.1 8B, Aggregation=Exc, threshold=τ=0.92026.0539.3