{"arxiv_id":"2608.12307","incoming":{"papers":1,"sentences":1,"groups":[{"kind":"background","label":"Cites it as background","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2608.27266","title":"Naive Prompt Optimization: Rethinking the Need for Complex Prompt Search","url":"/paper/2608.27266","section":"Introduction","sentence":"AI4AI [ref] studied how a teacher model builds an inference-time harness for weaker student models.","intent_p":0.99,"contests_p":null}]}],"unlabelled_sentences":0},"outgoing":{"papers":15,"sentences":18,"groups":[{"kind":"method_or_tool","label":"Uses the method or tool of","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2605.27721","title":"UserHarness: Harnessing User Minds for Stronger Agent Theory-of-Mind","url":"/paper/2605.27721","section":"Experimental Setup","sentence":"This setting yields a macro-average accuracy of $$ for GPT-5.4-mini and $$ for Gemini-3.5-flash. * Human-Inspired Harness: Each target model is evaluated with UserHarness [ref] , a human-designed harness framework for ToM problems.","intent_p":0.86,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":6,"sentences":7,"quotes":[{"arxiv_id":"2203.11171","title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","url":"/paper/2203.11171","section":"Related Work","sentence":"Chain-of-thought prompting elicits intermediate reasoning steps from large models [ref] , self-consistency improves reliability by sampling multiple reasoning paths and marginalizing over their answers [ref] , and least-to-most prompting decomposes hard problems into easier subproblems solved sequentially [ref] .","intent_p":0.96,"contests_p":null},{"arxiv_id":"2205.10625","title":"Least-to-Most Prompting Enables Complex Reasoning in Large Language Models","url":"/paper/2205.10625","section":"Related Work","sentence":"Chain-of-thought prompting elicits intermediate reasoning steps from large models [ref] , self-consistency improves reliability by sampling multiple reasoning paths and marginalizing over their answers [ref] , and least-to-most prompting decomposes hard problems into easier subproblems solved sequentially [ref] .","intent_p":0.96,"contests_p":null},{"arxiv_id":"1503.02531","title":"Distilling the Knowledge in a Neural Network","url":"/paper/1503.02531","section":"Introduction","sentence":"Recent progress in model distillation has made it increasingly plausible to deploy smaller language model experts in settings that once required much larger ones [ref] .","intent_p":0.87,"contests_p":null},{"arxiv_id":"2211.12588","title":"Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks","url":"/paper/2211.12588","section":"Related Work","sentence":"Program-aided methods such as PAL and Program-of-Thoughts ask the model to translate a problem into executable code, leaving exact computation to a Python interpreter [ref] .","intent_p":0.94,"contests_p":null},{"arxiv_id":"2210.02406","title":"Decomposed Prompting: A Modular Approach for Solving Complex Tasks","url":"/paper/2210.02406","section":"Related Work","sentence":"Decomposed prompting generalizes this modular view by delegating subproblems to specialized prompts, models, or symbolic functions [ref] .","intent_p":0.81,"contests_p":null}]}],"unlabelled_sentences":10},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2608.12307"}}