Ceiling

Follow this topic Ask about this topic

Momentum

9 papers in the last four weeks, up 200% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 44

All topics

Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost
Sep 30, 2026Sietse SchelpeCeiling
Corbenic AI
You Cannot Recover What Was Never Measured: Quantifying the Information Ceiling of Ultra-Low-Field MRI Super-Resolution
Sep 29, 2026Prathamesh Pradeep Khole, Shreya Handa, Utkarsh Gupta +1Magnetic Resonance Imaging Generative Super-Resolution
University of California, Santa Cruz · Johns Hopkins University
Feedback Makes Perfect: A Closed-Loop Framework for NL-to-STL Translation
Sep 27, 2026Bowen Ye, Xiang YinSignal Temporal Logic Natural Language
Shanghai Jiao Tong University Shanghai, China
The Copy Ceiling: An Input-Exposure Control for Ontology-Grounded Generation over Curated Corpora
Sep 21, 2026John J. O'HareText Corpora Exposure
DreamLab AI, The Loom / VisionFlow neurosymbolic stack
The Organization of Inference: Information, Resource Constraints, and AI Production
Sep 17, 2026Yukun Zhang, Kemu Xu, Yishen ChenAgentic Inference Artificial Intelligence Systems
The Chinese University of Hong Kong Hong Kong, China · University of Edinburgh Edinburgh, United Kingdom · The Chinese University of Hong Kong, Shenzhen Shenzhen, China
The segmentation ceiling: why explicit left-ventricular masks do not improve learned ejection-fraction regression
Sep 17, 2026Farshid Farhadi Khouzani, Paul La Plante, Bryar Mustafa Shareef +1Left-Ventricular Ejection Fraction Ceiling
Department of Computer Science, University of Nevada, Las Vegas, Las Vegas, NV, USA
Empirical Evaluation of Task-Based Permission Scoping Architecture for AI Agents
Sep 14, 2026Halil Burak NoyanAccess Control Kernel-Resident Tool Governance Gateway
Independent Researcher
What an odour descriptor corpus can and cannot measure: valence, attenuation, and the ceiling of the public record
Sep 14, 2026Stylianos Kampakis, Fabio RovaiSmells Inter-Annotator Agreement
Tesseract Academy, London, UK
Breaking the Token Ceiling: Distilling Smaller, Stronger Byte Models
Sep 14, 2026Kalyani Marathe, Artidoro Pagnoni, Tomasz Limisiewicz +4Small Large Language Models Transformer Encoder
University of Washington, Seattle · Work done at Meta FAIR · Meta FAIR
Geometric Ceilings on Time-Frequency Masking for Single-Channel Separation
Sep 3, 2026Maxime BaeldeSpeech Separation Time-Frequency
Independent researcher, Lille, France
Separating perception from reasoning in vision-language models: a model-free render ceiling for crystal structures
Sep 1, 2026Can Polat, Mustafa Kurban, Erchin Serpedin +1Recent Vision-Language Models Ceiling
Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX, USA · Department of Electrical and Computer Engineering, Texas A&M University at Qatar, Doha, Qatar · Department of Prosthetics and Orthotics, Ankara University, Ankara, Turkey +1
Aerial Layouting: Design and Control of a Compliant and Actuated End-Effector for Precise In-flight Marking on Ceilings
Aug 11, 2026Christian Lanegger, Marco Ruggia, Marco Tognon +2Aerial Robots End-Effector Pose
Autonomous Systems Lab, ETH Zurich, Switzerland
Floor, Ceiling, and the Fusion Gap: How Much of Crowd Reading Attention Can Machines Predict?
Aug 3, 2026Kazuki Nakayashiki, Keisuke WatanabeReadability Ceiling
Glasp Inc.
The Label Defines the Timescale: Trait-State Limits of Temporal-Aggregate Learning
Aug 3, 2026Xizhe ZhangTimescale Training Time
Who Belongs in the Eval Set? A Capability-Taxonomy-Driven Pipeline for Curating Regression Eval Sets in Agent-Extensibility Platforms
Aug 2, 2026Tezan Sahu, Aritra Das, Pankaj Mittal +1Agentic Evaluations Data-Curation
Microsoft Hyderabad, India · Indian Institute of Technology Roorkee Roorkee, India
The Sparsity Ceiling: Where Spiking Networks Can and Cannot Trade Activity for Energy
Jul 29, 2026Zeyu WangSpiking Neural Networks Neuromorphic Computing
Georgia Institute of Technology
Tripody: An Overconstrained 3-SPR-like Parallel Robot for High-Reach Construction Tasks
Jul 28, 2026Julien Kindle, Jakub Raczy, Riccardo Balbi +3Robotic Manipulator Robot Systems
Robotic Systems Lab, ETH Zurich, Switzerland · Hilti AG, Schaan, Liechtenstein
Are You Still the Agent I Authorized? Earned Authority under a Fixed Ceiling for Evolving Agents
Jul 26, 2026Zhaoxi Zhang, Xiaomei ZhangAuthorization Authority
University of Technology Sydney · Griffith University
Dynamic Capability Scoping for Enterprise AI Agents: A Synthetic Dataset and Three-Source Permission Architecture
Jul 24, 2026Halil Burak NoyanIntegrity Synthetic Data
Independent Researcher, London, United Kingdom.
FORGE-plus: Force-Budgeted Recovery for Contact-Rich Assembly with a Frozen LLM Supervisor
Jul 23, 2026Kyupaeck Jeff Rah, Midum OhGoal-Conditioned Reinforcement Learning Slipsense
Independent Researcher
Position Bias is Hidden Behind Ceiling Effects: A Permutation Diagnostic for LLM Benchmarks
Jul 23, 2026Hiroki TambaPosition Bias Large Language Model Evaluation
Independent researcher.
Routing Ceilings Are Domain-Independent: Structural Prior Injection in Code Security Vulnerability Detection
Jul 16, 2026Manuel Israel CázaresCommon Vulnerabilities And Exposure Code Quality
Bytepro AI Mazatlán, Sinaloa, Mexico
Partially Correlated Verifier Cascades in LLM Harnesses: Concave Log-Odds, Polynomial Reliability, and Blind-Spot Ceilings
Jul 15, 2026Jiangang HanVerification Framework Finite-Sample Certificates
Independent researcher
From Text to Parameters: Predicting Item Parameters from Embedding Regularization with Reliability and Design Ceilings
Jul 8, 2026Shi-Ting Chen, Jinsong ChenPsychometric Properties Logistic Regression
Faculty of Education, The University of Hong Kong
Computing with Stochastic Oracles in AI-Augmented Computation
Jul 8, 2026Jie WangFirst Order Oracle Complexity Counter Turing Test
Richard A. Miner School of Computing and Information Sciences, University of Massachusetts, Lowell, MA 01854, USA.
World Feedback for Clinical Agents: Diagnosing RL in FHIR Environments
Jul 1, 2026Ananya Mantravadi, Harshit Rajgarhia, Prasanna Desikan +1Icr-Rl Memoryagentbench
Centific Global Solutions, Inc.
Anchoring on Reality: Breaking the Pseudo-Target Ceiling in Makeup Transfer
Jun 30, 2026Bo Wei, Xianhui Lin, Yi Dong +8Synthetic-To-Real Domain Gap Photorealistic Style Transfer
1Harbin Institute of Technology · 2vivo BlueImage Lab, vivo Mobile Communication Co., Ltd · 3Nanjing University
When More Sampling Hurts: The Modal Ceiling and Correlation Ceiling of Test-Time Scaling
Jun 27, 2026Yong Yi Bay, Kathleen A. YearickTest-Time Scaling Reasoning Benchmark
University of Illinois at Urbana-Champaign
JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting
Jun 16, 2026Lanxiang Hu, Zhaoxiang Feng, Yulun Wu +9Speculative Decoding Block-Diffusion Drafter
UC San Diego · Zhejiang University · UIUC +2
LoHoSearch: Benchmarking Long-Horizon Search Agents Beyond the Human Difficulty Ceiling
Jun 11, 2026Jiarui Zhao, Rongzhi Zhang, Lingchuan Liu +3Search Agents Question-Answering Benchmarks
Meituan
QO-Bench: Diagnosing Query-Operator-Preserving Retrieval over Typed Event Tuples
Jun 3, 2026Mengao Zhang, Xiang Yang, Chang Liu +2Question-Answering Benchmarks Diagnostic Benchmark
Nanyang Technological University · Asian Institute of Digital Finance, National University of Singapore
A Two-Phase Stability Study of LLM Judges and Bar Council Examiners on Thai Bar-Exam Free-Form Essays
May 25, 2026Pawitsapak Akarajaradwong, Wuttikrai Lertprasertphakorn, Chompakorn Chaksangchaichot +1Llm-As-A-Judge Large Language Model Judges
VISAI AI · Independent Researcher · Vidyasirimedhi Institute of Science and Technology
Acting on the Unseen: Communication-Free Collaborative Filtering for Decentralized Multi-Robot Task Allocation
May 25, 2026Alexander Apartsin, Yigal Meshulam, Yehudit ApersteinMulti-Robot Systems Collaborative Filtering
Holon Institute of Technology, Holon, Israel. · Afeka Tel Aviv Academic College of Engineering, Tel Aviv, Israel
Beyond Inference-Only Deployment: Comparing Weight-Based Consolidation Against Cascading Compaction
May 23, 2026Simon Dennis, Kevin Shabahang, Hao Guo +1Compaction Llm-As-A-Judge
i14 / University of Melbourne · i14
Training distribution determines the ceiling of drug-blind cancer sensitivity prediction
May 20, 2026Taekyung HeoPharmacokinetics Cells
Do Language Models Align with Brains? Prediction Scores Are Not Enough
May 13, 2026Xiao JiaLarge Language Model Alignment Language Modeling
School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen, China
Decomposing the Generalization Gap in PROTAC Activity Prediction: Variance Attribution and the Inter-Laboratory Ceiling
May 12, 2026Thor Klamt, Wolfgang Nejdl, Ming TangProtein Receiver Operating Characteristic Area Under The Curve
L3S Research Center, Leibniz Universität Hannover, Appelstraße 9a, 30167 Hannover, Germany · Institute of Data Science (Knowledge-Based Systems), Faculty of Electrical Engineering and Computer Science,2026 Leibniz Universität Hannover, Appelstraße 9a, 30167 Hannover, Germany
Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion
May 12, 2026ShiYing Huang, Liang Lin, Yuer Li +6Preference Alignment Multi-Objective Reinforcement Learning
Huazhong University of Science and Technology · Nanyang Technological University · Tsinghua University +1
Agentic AIs Are the Missing Paradigm for Out-of-Distribution Generalization in Foundation Models
May 7, 2026Xin Wang, Haibo Chen, Wenxuan Liu +1Wireless Foundation Models Out-Of-Distribution
Tsinghua University
Raising the Ceiling: Better Empirical Fixation Densities for Saliency Benchmarking
May 5, 2026Susmit Agrawal, Jannis Hollman, Matthias KümmererSaliency Density Ratio Estimation
Tübingen AI Center, University of Tübingen, Germany
MAD-OPD: Breaking the Ceiling in On-Policy Distillation via Multi-Agent Debate
May 2, 2026Jianze Wang, Ying Liu, Jinlong Chen +7Aware Heterogeneous Multi-Teacher Multimodal On-Policy Distillation Token-Level Supervision
School of Artificial Intelligence and Automation, Huazhong University of Science and Technology · Alibaba Group
The signal is the ceiling: Measurement limits of LLM-predicted experience ratings from open-ended survey text
Apr 21, 2026Andrew Hong, Jason Potteiger, Luis E. ZapataSurvey Ceiling
Dimension Labs
Less Is More: Cognitive Load and the Single-Prompt Ceiling in LLM Mathematical Reasoning
Apr 20, 2026Manuel Israel CazaresMathematical Reasoning LLM Reasoning Strategies
Bytepro AI Mazatlán, Sinaloa, Mexico
CeRA: Breaking the Linear Ceiling of Low-Rank Adaptation with Non-linearity Retained at Inference
Feb 26, 2026Hung-Hsuan ChenLow-Rank Adaptation Framework Rapid Adaptation
Computer Science and Information Engineering National Central University Taoyuan, Taiwan

Sep 30, 2026, cs.CLJ/K move · Enter open · S save

Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost

Sietse Schelpe

Corbenic AI

Ceiling

Open paper AskPDF arXiv

A transformer language model performs a bounded amount of computation per token, and recent work by Vishal Sikka, former CEO of Infosys, argues that this bound limits which tasks a model can carry out or verify (arXiv:2507.07505). We ask how much of the budget beneath that ceiling is spent on work the model has already done. Serving is stateless across requests: a model that answers a second question about a document recomputes the document's attention state from the first token. On seven real-world datasets, 98.7% of prompt tokens were text the model had already read. We present Galahad, a memory layer for vLLM, SGLang and llama.cpp that makes this reading a one-time cost. Taliesin saves the model's key-value (KV) state for a block of text and loads it on the next request that contains the same bytes, instead of recomputing it. Blaise keeps the documents themselves and passes the model only the section a question needs. On a recall test with 100 facts hidden in a 97,000-token corpus (Gemma 4 31B), Taliesin alone let the model attend to the whole corpus and answered 98 of 100 on llama.cpp at 3.0 s and 572 J per question, against 10 of 100, 9.3 s and 2,754 J for the same model without Galahad, which could hold only the last 12,000 tokens. With Blaise added, the model read about 668 tokens per question and answered 100 of 100 on all three runtimes at 0.59-0.64 s and 200-213 J; a tuned RAGFlow pipeline answered 77. Storing the corpus is a one-time cost of about 100 s and 28 kJ, whose energy is recovered after 13 questions. Restored state is bit-identical: all 262,144 output logits matched after restart, rehydration and hot-load. Galahad worked with all 30 models we tested under vLLM, and it fails closed: any load that does not pass its checks is recomputed. Together these results move LLM serving from stateless to stateful inference.

Momentum

Related topics

Latest papers 44