Specialty
A replay-free method for continual learning scored higher across the reported benchmarks and showed no measured retention loss on TRACE, though its routing has not been tested under severe distribution shifts.
Scientific publication4 min read
A preprint describes TACD, a distillation method that limits terminal supervision and reports stronger benchmark results, faster generation and lower peak GPU memory across text-to-motion tests.
Scientific publication4 min read
Camera-based AI scoring closely matched consensus video review for total BESS scores and produced smaller discrepancies than real-time on-site manual scoring in 80 participants.
Sports Medicine - Open4 min read
A reinforcement-learning retriever called MO-IKE reported higher combined scores than DR-IKE on Llama and Mistral tests, and recorded an overall score of 24.5 on UniEdit.
Scientific publication4 min read
Thirteen frontier models were tested on 20 computational-biology tasks, and none scored above 0.48 overall; performance was lower on very large inputs and deep analysis chains.
Scientific publication4 min read
A literature-based review proposes an iterative workflow for AI systems that carry out multi-step data science, while stressing that the framework has not been experimentally validated.
Scientific publication4 min read
A preprint study found that the top-ranked token changed a model’s reported credal width more than random eligible tokens, while numerical checks exposed limits in the attribution values.
Scientific publication4 min read
A benchmark of language-model agents found that strong performance on static enterprise questions did not reliably identify the strongest systems in simulated interactive tasks.
Scientific publication4 min read
A computational study of 10 LLM backbones found that SimpleEvol had the lowest paper-defined framework-handcraftedness score and the highest reported intelligence-conversion efficiency on two primary tasks. Its direct GPT-5-mini results also led every listed TSP and CVRP setting, but fitted slopes turned negative on Taillard out-of-distribution instances.
Scientific publication4 min read
A preprint describes RouteFM, a frozen language-model router that uses past model behavior to make choices in new environments, and reports the highest routing quality across its held-out and multimodal tests.
Scientific publication3 min read
A value-of-information policy called REVOIR was designed to decide when asking a question is worth the effort, and benchmark results reported fewer questions alongside higher correctness.
Scientific publication3 min read
An arXiv preprint reports that Veer recorded fewer deceptive outcomes than matched No-Defense runs across controlled benchmark tests while retaining task success.
Scientific publication4 min read
A preprint reports lower energy-forecast error across 18 test sets and lower reported battery-decision regret and operating costs in one UK case.
Scientific publication3 min read
Tests on a frozen language model recorded sharply lower prompt-injection rates with a learned overlay, while other overlay-controlled tasks showed uneven results.
Scientific publication4 min read
A preprint tests a method for reading factual or behavioral information from frozen model updates and using that signal to target sparse interventions. It found measurable alignment in token-level loss and selected downstream comparisons, but complete descriptions of unseen updates remained uncommon.
Scientific publication4 min read
A preprint reports that SAPO matched or slightly exceeded PPO on ALFWorld and WebShop while using less peak GPU memory and less time per training iteration.
Scientific publication4 min read
An arXiv preprint reports higher math and code benchmark averages for a reward-aligned trajectory filter, with total training time close to standard on-policy distillation.
Scientific publication4 min read
A preprint presents AERA, a controller that estimates whether more reasoning may help and reports near-baseline GSM8K accuracy with sharply lower generation use.
Scientific publication5 min read
An open-source harness for long-horizon coding agents scored above selected systems on two benchmarks, but the study did not isolate the harness from model, prompt, tool and implementation differences.
Scientific publication4 min read
A method called DA3PO was reported to outperform GRPO, DAPO and GSPO on mathematical reasoning benchmarks, though the tests covered only two Qwen3 base models and did not report uncertainty estimates.
Scientific publication4 min read
A preprint describes an AI system that retrieves evidence before choosing how its agents collaborate. It reports higher scores across seven reasoning benchmarks, but the evidence comes from in-silico evaluations rather than human users or real-world deployments.
Scientific publication4 min read
A preprint reports that GOD linked browser commands to execution records, replay checks and portable packs in 15 completed run slots. The evaluation focused on command, replay and artifact checks, not human or social validity.
Scientific publication4 min read
A preprint examining French and Egyptian Arabic movie dialogue finds that six AI systems align more closely with humans on visible social cues than on subtle power relationships, while multimodal results are limited by incomplete coverage.
Scientific publication4 min read
The model with the highest pooled scores retained minute-level overnight patterns, but the small hospital sample and limited calibration do not support individual care use.
Scientific publication4 min read
The paper presents CrabOS as an operating-system approach to human-AI handoffs, using shared text objects and common capability controls, but reports architecture and case studies rather than measured evidence of better results.
Scientific publication4 min read
A text-only reconstruction of Reactome preserved its large-scale network pattern, but the study measured graph similarity rather than biological correctness.
Scientific publication3 min read
The preprint reports perfect syntax across the tested settings, while task-accuracy scores for SQL and linear temporal logic remained uneven.
Scientific publication4 min read
A benchmark study found the largest routing gap when user-profile details changed which skill best matched a task, while profile-insensitive cases stayed similar.
Scientific publication3 min read
A preprint reports stronger image-quality results from Flow Matching than diffusion on sparse-view chest CT, with FlowDPS leading the reconstruction comparison.
Scientific publication4 min read
A diagnose-then-correct system called DoCtOR reported higher success rates on HotPotQA, ChartQAPro and Mind2Web. Its attribution model was less accurate on held-out failures, and open-ended, multilingual and larger-team settings were not tested.
Scientific publication4 min read
A buffered sample-selection method called RECAST reported lower blood-pressure estimation error than TTC and no adaptation on PulseDB and MC-MED. The results came from simulated streaming benchmarks, with extra but sub-second latency and no prospective clinical evaluation reported.
Scientific publication5 min read
A methods preprint compares three real-valued position encodings, finding similar classification averages and an exact, faster algebraic shift for its Sinusoid variant.
Scientific publication4 min read
GRACE, a gradient-guided method for selecting compact forget and retain sets, reported higher forget-sample retrieval accuracy than RASLIK and the highest model utility in six of eight algorithm-model combinations, while forget-quality scores varied little across selectors.
Scientific publication4 min read
A randomized experiment found that list position still shaped which hotels AI agents inspected, but much less than in a human benchmark; effects on final booking varied by model and were no longer statistically significant at the highest tested reasoning effort.
Scientific publication4 min read
A computational preprint reports sharply lower code and math scores in targeted MLP-neuron suppression tests, while several non-target benchmarks retained much higher scores.
Scientific publication5 min read
ProgRouter reported strong task and citation results while tracking operating-cost budgets, GPU energy use, and execution time across four benchmark streams.
Scientific publication3 min read
A study of 536 failed web-task runs found that whole-trajectory reruns often produced different outcomes without showing that the recorded failure mechanism had been corrected.
Scientific publication4 min read
A computational evaluation reports that HarnessLens improved average held-out performance by 7.6% to 13.6% across three agent harnesses and four benchmarks, while using a smaller configured interaction budget than the baselines.
Scientific publication5 min read
A preprint reports that KLOD can update targeted facts inside language models while keeping edit success near 100 percent in benchmark tests, although broader generalization and preservation of unrelated predictions remain in tension.
Scientific publication5 min read
A training method for autoregressive spiking language models improved reported benchmark averages over matched knowledge-distillation checkpoints and stayed non-collapsed in a 10-seed stress test. The work remains an early computational evaluation, with no hardware measurements or evidence about human language performance.
Scientific publication5 min read