WARC learns unseen game mechanics from limited test-time exposures
The 35M-parameter model predicts four in five held-out mechanics at 80% cell accuracy; half reach that within about 80 exposures.
Stories per day · 30 days
2 RSS
Relatedcompression2kv-cache2long-context2qwen32ruler2arc-agi1efficiency1game-mechanics1language-models1mamba1memory1minicpm1minicpm51open-weight-models1recurrent-models1scientific-discovery1world-models1
11 stories, newest first · one card per story
Stories tagged #test-time-training
ParaTTT parallelizes deep nonlinear test-time training memories
Two refinement steps at 4K context evaluate the inner-loop trajectory 32× faster than TTT-MLP at matched error.
Test-time memory evolution beats test-time training on math discovery
TTME updates external memory instead of weights and cuts FLOPs by 82.8–84.6% and runtime by 47.7–58.4%.
Latent Compaction with Test-Time Training
Latent Compaction preserves long-context retrieval at 32x KV-cache compression via test-time training.
Latent Compaction with Test-Time Training
Latent Compaction keeps 96.7% RULER accuracy at 32x KV-cache compression via test-time training.
ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified Experience
ASCENT updates deployed LLM agents by self-distilling verified trajectories into LoRA weights at test time.
Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation
Learning from self-generated text destabilizes test-time training across 128K tokens, including Adam updates on Qwen3-4B.
JEPA-TTT: Persistent Test-Time Training of Latent World Models for Planning under Dynamics Shifts
JEPA-TTT adapts a pretrained latent world model at test time when environment dynamics shift.
Breaking Babel: A Self-Evolving Multi-Agent System for Long-Form Subtitle Translation
SMART, a self-evolving multi-agent translator, tops Subtitle Arena and MuSC by adapting prompts and routing at test time.
Using Context Is Not Enough: Test-Time Training for Personalized Reward Modeling
P-TTT personalizes reward models with test-time fast weights, beating in-context preference conditioning without backprop.
Distill Globally, Adapt Locally: Reasoning Distillation and Product-Type Test-Time Training for Scalable Trade-Up Recommendation
A distillation framework compresses LLM reasoning into a 15.5M-parameter trade-up recommendation model reaching AUC 0.941 with product-type test-time training.