Upload evidence_capsules.jsonl

capsules/evidence_capsules.jsonl

@@ -0,0 +1,9 @@
+{"capsule_id": "E_REACT_2210.03629", "source_url": "https://arxiv.org/abs/2210.03629", "excerpt": "Generate reasoning traces and task-specific actions in an interleaved manner; actions interface with external sources to gather additional information.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_RAG_2005.11401", "source_url": "https://arxiv.org/abs/2005.11401", "excerpt": "Retrieval-augmented generation combines parametric and non-parametric memory for language generation, using a dense vector index accessed with a retriever.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_COT_2201.11903", "source_url": "https://arxiv.org/abs/2201.11903", "excerpt": "Generating a chain of thought—intermediate reasoning steps—significantly improves large language models on arithmetic, commonsense, and symbolic reasoning tasks.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_SELFCONS_2203.11171", "source_url": "https://arxiv.org/abs/2203.11171", "excerpt": "Self-consistency samples diverse reasoning paths then selects the most consistent answer by marginalizing out sampled paths, improving chain-of-thought decoding.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_SELFREFINE_2303.17651", "source_url": "https://arxiv.org/abs/2303.17651", "excerpt": "Iterative feedback and refinement: generate initial output, then the same LLM provides feedback and uses it to refine, without supervised training.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_REFLEXION_2303.11366", "source_url": "https://arxiv.org/abs/2303.11366", "excerpt": "Reinforce language agents through linguistic feedback: reflect on task feedback signals and keep reflective text in episodic memory to improve later trials.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_DSPY_2310.03714", "source_url": "https://arxiv.org/abs/2310.03714", "excerpt": "DSPy abstracts LM pipelines as text transformation graphs and offers a systematic approach to develop and optimize pipelines beyond hard-coded prompt templates.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_GUARDRAILS_REASK", "source_url": "https://www.guardrailsai.com/docs/how_to_guides/rail", "excerpt": "Reask: regenerate output that meets quality criteria; the reask prompt includes which criteria failed, auto-generated by the validator.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}
+{"capsule_id": "E_LANGGRAPH_MEMORY", "source_url": "https://docs.langchain.com/oss/python/langgraph/memory", "excerpt": "Short-term memory tracks ongoing conversation by maintaining message history; state persists via a checkpointer and is read at the start of each step.", "license_note": "Short excerpt from abstract/docs (keep <=25 words when publishing; expand via links)."}