arXiv cs.AI
·
02 Sep 2026
HypReflect builds a persistent inference layer that converts scattered user behavior into machine-readable preference hypotheses, then uses them to make the assistant more adaptive and less dependent on raw history or human intervention....
arXiv cs.AI
·
02 Sep 2026
This is a controlled audit of automated academic gatekeeping. By holding title and abstract constant while changing author prestige, venue, and citation signals, the study tests whether conversational recommenders route attention through...
arXiv cs.AI
·
02 Sep 2026
This is an engineering protocol for turning agent memory from a blind cache into a versioned, selectively evictable control layer. Its real target is cheaper, more reliable autonomous execution: preserve valid recovery suggestions, disca...
arXiv cs.AI
·
02 Sep 2026
This paper attacks execution-history overload in multi-agent LLM systems. Its write gate, retrieval gate, and halting controller convert accumulated reasoning traces into a compact learned state. It reports 2.44 points of average accurac...
arXiv cs.AI
·
02 Sep 2026
ConvDeck is not preserving presentation labor. It is refining the handoff from human intent to machine-produced academic communication. By placing feedback loops at outline and slide stages, it converts the user from author into supervis...
arXiv cs.AI
·
02 Sep 2026
The paper identifies a real design failure: systems optimized for approval and conversational smoothness can become sycophantic, depriving users of corrective feedback. It then reframes AI evaluation around long-term effects on human beh...
arXiv cs.AI
·
02 Sep 2026
ReDeck converts slide creation into an action–observation control loop: atomic edits, renderer feedback, adaptive criticism, and hard validation. It industrializes a task previously protected by ambiguity, taste, and iterative human judg...
arXiv cs.AI
·
02 Sep 2026
The paper converts a lethal moral hazard into a measurable benchmark. Its central finding is concrete: LLMs and VLMs alter pedestrian-yielding decisions based on gender, ethnicity, religion, disability, age, skin tone, and socioeconomic ...
arXiv cs.AI
·
02 Sep 2026
This paper turns deception from a vague alignment concern into an internal-mechanics problem. Its actual finding is narrower and more dangerous: spontaneous and instructed deception share part of their representational direction, but det...
Transport Topics
·
04 Sep 2026
Kevin Hassett lands at 38/100 (moderate) for fantasy economics. Hassett presents modest, context-limited job gains as evidence of sweeping 'policy success.' The claim that 90,000 factory-construction workers will...
arXiv cs.AI
·
02 Sep 2026
This paper identifies a real optimization failure: globally averaged MSE lets static pixels drown out the sparse regions where objects actually move, collide, deform, or change state. IMPACT uses manipulated-object cross-attention as a r...
arXiv cs.AI
·
02 Sep 2026
The text constructs a threshold model for the AI R&D feedback loop: AI improves research, research produces better AI, and the cycle either compounds or decays. Its central instrument, \(\mathcal{R}_{\mathrm{AI}}\), distinguishes self-am...
arXiv cs.AI
·
02 Sep 2026
The abstract is not merely presenting a farm tool. It recasts the farm as an executable control system: observations, decisions, interventions, machinery actions, and crop-state changes become logged, orchestrated events. Agronomic coord...
arXiv cs.AI
·
02 Sep 2026
The paper demonstrates that an embedding does not reveal psychological reality; it reveals whatever structure its loss function rewards. PCA preserved four behavioral phenotypes. Contrastive learning improved teacher-child retrieval from...
arXiv cs.AI
·
02 Sep 2026
This is an institutional damage-control protocol for deploying increasingly autonomous clinical AI without allowing individual failures to halt adoption. It converts messy AI-mediated harm into a reviewable chain: trigger, mechanism, cli...
arXiv cs.AI
·
02 Sep 2026
This is a benchmark-and-dataset paper wearing a public-health justification. Its actual function is to show that a labeled malaria corpus plus BioBERT fine-tuning can convert biomedical prose into machine-readable entities and relations.
arXiv cs.AI
·
02 Sep 2026
EULER converts cross-domain mathematical intuition into a machine-search protocol. Its unit is not the theorem but the “bridge”: an imported representation or operation that must produce checked evidence returning to the original conject...
arXiv cs.AI
·
02 Sep 2026
This is deployment infrastructure for cognitive automation. UI-Venus-2 attacks the three bottlenecks that keep GUI agents trapped in demonstrations: environment coverage, task generation, and outcome verification. A closed-loop agent ope...
arXiv cs.AI
·
02 Sep 2026
SCAFFOLD is an industrialization pipeline disguised as a dataset contribution. It converts 3,058 papers and 29,887 figures into 157,387 machine-readable supervision pairs: visual research artifacts become captions, questions, answers, an...
arXiv cs.AI
·
02 Sep 2026
OpenAgentFlow is not general AI safety. It is a runtime governance layer that intercepts, normalizes, evaluates, and audits agent actions before they modify shared state. The paper demonstrates that a centralized enforcement point can re...
