{"paper":{"title":"Reasoning or Just Prompting? Probing the Mechanism of Latent Chemical Reasoning","arxiv_id":"2610.34173","primary_cat":"cs.LG","submitted_at":"2026-10-25T09:03:36+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Latent reasoning performs intermediate computation in continuous space rather than through explicit tokens, which is particularly appealing in chemistry, where errors in explicit reasoning can propagate to the final answer.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.11,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.78,"id":"s1","own":false,"role":"gap","text":"Yet if latent states truly constitute an implicit reasoning chain, they should be causally important.","hedge":{"p":0.48,"label":"hedged_possibility"},"own_p":0.05,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.78,"method":0.0,"result":0.0,"background":0.22,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":1.0,"id":"s2","own":true,"role":"result","text":"We find otherwise: swapping or moderately perturbing intermediate latent states has little effect on performance, while removing them entirely sometimes causes a substantial score drop.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.86,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.71,"id":"s3","own":true,"role":"interpretation","text":"Thus, the latent states matter, but their molecule-specific content appears to matter far less.","hedge":{"p":0.82,"label":"asserted"},"own_p":0.92,"main_claim_p":0.16,"quantitative":true,"quantitative_p":0.88,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.29,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.71}},{"i":4,"p":0.99,"id":"s4","own":false,"role":"interpretation","text":"This points to a simpler interpretation.","hedge":{"p":0.78,"label":"unclear"},"own_p":0.12,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.01,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.99}},{"i":5,"p":1.0,"id":"s5","own":false,"role":"interpretation","text":"Rather than encoding a step-by-step reasoning chain, latent states may primarily steer subsequent model computation, much like soft prompts.","hedge":{"p":1.0,"label":"hedged_possibility"},"own_p":0.23,"main_claim_p":0.43,"quantitative":false,"quantitative_p":0.15,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}},{"i":6,"p":0.97,"id":"s6","own":true,"role":"result","text":"Indeed, a single fixed prefix shared across molecules reproduces LatentChem's performance, and a learned static soft prompt performs comparably to per-input latent generation in controlled models.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.88,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.9,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.97,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.02}},{"i":7,"p":0.73,"id":"s7","own":true,"role":"interpretation","text":"The tolerance of latent sequences to substitution and moderate perturbation further supports this connection.","hedge":{"p":0.52,"label":"asserted"},"own_p":0.83,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.1,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.27,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.73}},{"i":8,"p":1.0,"id":"s8","own":true,"role":"interpretation","text":"We therefore suggest that latent chemical reasoning may be better viewed as soft prompting than as a reasoning chain hidden in continuous space.","hedge":{"p":1.0,"label":"hedged_possibility"},"own_p":0.57,"main_claim_p":0.93,"quantitative":false,"quantitative_p":0.49,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}}]},"compiled":{"main_claim":{"p":1.0,"text":"We therefore suggest that latent chemical reasoning may be better viewed as soft prompting than as a reasoning chain hidden in continuous space.","location":"abstract s8","hedge_strength":{"p":1.0,"label":"hedged possibility"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":6733,"cost_usd_reported":0.0},"compiled_at":"2026-10-27","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.03,"states_limitation":0.35,"human_participants":0.03,"controlled_comparison":0.84,"randomized_assignment":0.02,"states_quantitative_result":0.07,"compares_with_prior_results":0.67,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":0.83,"value":"observation or experiment","probabilities":{"argument":0.01,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.83,"simulation_or_computation":0.16}},"contribution_type":{"p":0.95,"value":"empirical study","probabilities":{"review":0.0,"method_or_tool":0.0,"empirical_study":0.95,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.05}},"contribution_statement":{"p":0.54,"text":"We find otherwise: swapping or moderately perturbing intermediate latent states has little effect on performance, while removing them entirely sometimes causes a substantial score drop.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":9,"sentences_in_abstract":9},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}