Continual-ARC scores lifelong skill acquisition on ARC-AGI streams
Harness persistence rarely beats a demonstration cache on memorization, and no harness method gains on unseen tasks, unlike weight updates.
Stories per day · 30 days
3 RSS
Relatedagent-harness1benchmark1benchmarks1continual-learning1epoch-ai1frontier-models1game-mechanics1gemini1google-deepmind1memorization1open-models1qwen1test-time-training1training-data1world-models1
12 stories, newest first · one card per story
Stories tagged #arc-agi
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.
arXiv studies training data for open models on ARC-AGI-2
Fine-tuning open models with specific rules and traces improves scores on ARC-AGI-2, with the best reaching 56.67.
Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months
Gemini 4 Argon scored 97.5% on ARC-AGI-3, near GPT-6 Astra but weaker in math and coding.
12
Scaling to Tens of Thousands of Test-Time Iterations with Loop-Native Attention Residuals
InfiLoop adds loop-native residuals so looped Transformers keep improving past 20,000 test-time steps.
Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
Memento 3 lets a frozen LLM learn explicit world models through a revisable natural-language rulebook.
Academia is for Ambition — Alex Zhang, MIT
MIT's Alex Zhang discusses Recursive Language Models, GPU kernels, and multi-agent harnesses.
VISTA: A Visual Harness for Reasoning in an Interactive World
VISTA's visual-memory harness lifts Claude Opus 5.0 to a perfect ARC-AGI-3 score using 57.4% fewer actions.
Schema: Discovering Unknown Environments via Agentic Program Induction
Schema agent harness lifts ARC-AGI-3 RHAE from 58.7% to 99.2% via program induction.
Allspark: Weak to Strong Transfer via Alternating Chain of Thought
Allspark transfers weak-model reasoning to stronger students via alternating chains of thought, including Kimi and Nemotron.
Recreating Minecraft Is Not a Benchmark
Opinion piece argues viral demos like one-prompt Minecraft recreations are overfit 'demo-benchmarks' measuring preparation, not true model capability.
2
GPT‑6 Astra
OpenAI launches GPT-6 Astra, a frontier model scoring 99.9% on ARC-AGI 3, priced $10/$50 per million tokens, rolling out to ChatGPT and API.