analiza aiqbench
This comprehensive survey synthesizes 312 papers to address reasoning hallucinations in large models, proposing a diagnostic framework (UIPCA) that tracks unsupported premises and invalid inferences throughout the model lifecycle. The analysis critically notes a current lack of empirical evidence demonstrating that specific interventions can simultaneously improve reasoning capabilities without compromising factual reliability under matched conditions. By connecting diagnosis to verification and selective release strategies, the work offers essential insights for managing persistent-state control in memory and tool usage within advanced AI systems.
Hallucination Across the Reasoning Lifecycle: Interface Visibility, Causal Evidence, and Release Control in Large Reasoning Models
arXiv:2610.05472v1 Announce Type: new Abstract: Reasoning errors can propagate into later decisions and memory. This survey synthesizes 312 papers and first-party reports on text-based reasoning hallucinations around three questions: what evidence is observable, what study designs establish, and which corrective actions the evidence supports. UIPCA records unsupported premises (U), invalid inferences (I), dependent reuse (P), visible answer-trace consistency (C), and action-policy failures (A). Across 58 reviewed sources, no comparison establishes that a specified intervention improves reasoning while reducing factual reliability under matched conditions. The synthesis connects diagnosis to verification, repair, selective release, and persistent-state control across memory, tools, and training feedback.
Abstract:Reasoning errors can propagate into later decisions and memory. This survey synthesizes 312 papers and first-party reports on text-based reasoning hallucinations around three questions: what evidence is observable, what study designs establish, and which corrective actions the evidence supports. UIPCA records unsupported premises (U), invalid inferences (I), dependent reuse (P), visible answer-trace consistency (C), and action-policy failures (A). Across 58 reviewed sources, no comparison establishes that a specified intervention improves reasoning while reducing factual reliability under matched conditions. The synthesis connects diagnosis to verification, repair, selective release, and persistent-state control across memory, tools, and training feedback.
35 pages, 8 figures, 9 tables. Electronic supplementary materials S1-S7 are included as ancillary files
Artificial Intelligence (cs.AI)
arXiv:2610.05472 [cs.AI]
(or arXiv:2610.05472v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.05472 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
supplement/CORPUS_PROVENANCE.md
supplement/FULLTEXT_FLOW_AUDITABILITY.json
supplement/FULLTEXT_FLOW_AUDITABILITY.md
supplement/README.md
supplement/S1_CORPUS_PROVENANCE.md
supplement/S1_manifest.csv
supplement/S1_manifest.md
supplement/S1_search_screening_manifest.csv
supplement/S1_search_screening_manifest.md
supplement/S2_corpus_catalog.csv
supplement/S3_consequential_claim_audit.csv
supplement/S3_external_27_criterion_adjudication.csv
supplement/S3_external_27_identity_and_retrieval.csv
supplement/S3_external_27_snippet_index.csv
supplement/S3_external_27_summary.json
supplement/S3_external_followup_priority.csv
supplement/S3_full_312_relevance_screen.csv
supplement/S3_fulltext_recheck_text_manifest.json
supplement/S3_original_34_sensitivity_recheck.csv
supplement/S3_screen_summary.json
supplement/S3_source_status_sensitivity.csv
supplement/S3_source_status_sensitivity.md
supplement/S3_strict_paradox_evidence.csv
supplement/S4_formal_definitions.md
supplement/S4_nearest_taxonomy_dual_coding.csv
supplement/S4_pilot_agreement.json
supplement/S4_pilot_agreement.md
supplement/S4_pilot_agreement.py
supplement/S4_pilot_alignment_changes.csv
supplement/S4_pilot_annotation_note.md
supplement/S4_pilot_dependency_review.csv
supplement/S4_pilot_disagreements.csv
supplement/S4_pilot_edges.csv
supplement/S4_pilot_records.csv
supplement/S4_pilot_sensitivity.json
supplement/S4_pilot_sensitivity.md
supplement/S4_pilot_sensitivity_changes.csv
supplement/S4_pilot_steps.csv
supplement/S4_pilot_summary.json
supplement/S4_pilot_unknown_rates.csv
supplement/S4_taxonomy_cases.json
supplement/S4_taxonomy_reference.py
supplement/S4_taxonomy_stress_cases.json
supplement/S4_taxonomy_stress_inputs.json
supplement/S5_method_benchmark_catalog.csv
supplement/S5_method_benchmark_catalog.md
supplement/S6_figure_data_and_build.csv
supplement/S6_figure_data_and_build.md
supplement/S7_research_river.pdf
supplement/S7_research_river_catalog.md
supplement/SEARCH_PROTOCOL.md
supplement/Supplementary_Methods.md
supplement/Supplementary_Methods.pdf
supplement/final_included_corpus.csv
supplement/final_local_evidence_manifest.csv
supplement/full_text_decisions.csv
supplement/full_text_to_corpus_reconciliation.csv
supplement/full_text_to_corpus_reconciliation_summary.json
supplement/incremental_20260830_flow_counts.json
supplement/incremental_20260830_fulltext_manifest.csv
supplement/incremental_20260830_include_release.csv
supplement/incremental_20260830_queries.json
supplement/incremental_20260830_query_counts.csv
supplement/incremental_20260830_screening_ledger.csv
supplement/incremental_20260830_search_hits.csv
supplement/input_manifest.json
supplement/memory_MAS_20260910_decisions.csv
supplement/memory_MAS_20260910_protocol.md
supplement/memory_MAS_20260910_queries.json
supplement/postreview_update_20260824.csv
supplement/queries.json
supplement/query_counts.csv
supplement/search_hits.csv
supplement/title_abstract_reconciliation.csv
supplement/uipca_pass_A.json
supplement/uipca_pass_B.json
supplement/verify_pilot_release.py
validation/CODEBOOK.md
validation/TUTORIAL_CASES.md
validation/VALIDATION_STATUS.md
validation/frozen/tutorial_case_manifest.csv
validation/sources/FaithCoT-BENCH/README.txt
validation/taxonomy_case_validation.json
validation/taxonomy_case_validation.md
validation/taxonomy_stress_validation.json
validation/taxonomy_stress_validation.md
validation/validate_taxonomy_cases.py
validation/validate_taxonomy_stress_cases.py
validation/verify_pilot_release.py
(85 additional files not shown) You must enabled JavaScript to view entire file list.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 308 cuvinte
Subiecte conexe

[AINews] not much happened today
Latent Space · 18 sept. 2026

Hands-free first notice of loss: Using Strands Agents and Amazon Bedrock AgentCore Browser Tool for intelligent claims intake
AWS · 9 iun. 2026
ServiceNow AI Control Tower: Selling a penthouse while building the frame
SiliconANGLE · 17 oct. 2026
Article: Your Next DSL Author is a Language Model
InfoQ · 16 sept. 2026