A study of seven LLMs finds that realistic prompt variations such as one-character misspellings trigger library hallucinations in up to 26% of cases, fabricated names in up to 99%, and time-based prompts in up to 85%, and introduces LibHalluBench for evaluation.
hub
Dated data: Tracing knowledge cutoffs in large language models
12 Pith papers cite this work, alongside 9 external citations. Polarity classification is still indexing.
12
Pith papers citing it
9
external citations · external index
hub tools
citation-role summary
background 2
method 1
citation-polarity summary
representative citing papers
PerGent, an agentic critique-refinement system for persona generation, reaches 96.9% expert approval in an industrial evaluation at Kinaxis and reproduces more pre-LLM expert content than single-shot baselines.
Many LLMs exhibit stronger environmental cognition, affect, and behavioral recommendations than human survey averages and shift with persona prompts.
Temporal knowledge drift is encoded as a geometrically orthogonal direction in LLM residual streams, independent of correctness and uncertainty.
AI coding agents and assistants fetch documentation in one or two HTTP requests with identifiable fingerprints, undermining standard web analytics.
PaST extracts a domain-agnostic skill vector from RL training and linearly injects it into SFT-adapted LLMs to improve knowledge use on QA and tool-use tasks.
CacheClip accelerates RAG prefill by up to 3.33x via auxiliary-model-guided selective KV recomputation while retaining 85-91% of full-attention quality on NIAH and LongBench.
MeMo encodes new knowledge into a separate memory model that integrates with frozen LLMs, showing strong performance on QA benchmarks while avoiding catastrophic forgetting and working without access to model weights.
TCFT trains LLMs on temporal critique tasks to reduce post-cutoff knowledge leakage by 37-42 percentage points over prompting and standard SFT on Qwen models.
LLM agent progress depends on externalizing cognitive functions into memory, skills, protocols, and harness engineering that coordinates them reliably.
ZoFia is a zero-shot fake news detection framework that uses hierarchical entity salience retrieval followed by multi-LLM adversarial debate to improve robustness over single-model approaches.
Mujica-MyGo decomposes multi-turn RAG interactions via multi-agent workflows and applies minimalist policy gradient optimization to improve performance on QA benchmarks while avoiding long-context problems.
citing papers explorer
-
Library Hallucinations in LLM-Generated Code: A Risk Analysis Grounded in Developer Queries
A study of seven LLMs finds that realistic prompt variations such as one-character misspellings trigger library hallucinations in up to 26% of cases, fabricated names in up to 99%, and time-based prompts in up to 85%, and introduces LibHalluBench for evaluation.
-
Agentic Persona Generation with Critique-Refinement: An Industrial Evaluation
PerGent, an agentic critique-refinement system for persona generation, reaches 96.9% expert approval in an industrial evaluation at Kinaxis and reproduces more pre-LLM expert content than single-shot baselines.
-
Greener Than Humans? Environmental Attitudes in Large Language Models
Many LLMs exhibit stronger environmental cognition, affect, and behavioral recommendations than human survey averages and shift with persona prompts.
-
The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations
Temporal knowledge drift is encoded as a geometrically orthogonal direction in LLM residual streams, independent of correctness and uncertainty.
-
Developer Experience with AI Coding Agents: HTTP Behavioral Signatures in Documentation Portals
AI coding agents and assistants fetch documentation in one or two HTTP requests with identifiable fingerprints, undermining standard web analytics.
-
Knowledge is Not Enough: Injecting RL Skills for Continual Adaptation
PaST extracts a domain-agnostic skill vector from RL training and linearly injects it into SFT-adapted LLMs to improve knowledge use on QA and tool-use tasks.
-
CacheClip: Accelerating RAG with Effective KV Cache Reuse
CacheClip accelerates RAG prefill by up to 3.33x via auxiliary-model-guided selective KV recomputation while retaining 85-91% of full-attention quality on NIAH and LongBench.
-
MeMo: Memory as a Model
MeMo encodes new knowledge into a separate memory model that integrates with frozen LLMs, showing strong performance on QA benchmarks while avoiding catastrophic forgetting and working without access to model weights.
-
Teaching Large Language Models When Not to Know: Learning Temporal Critique for Ex-Ante Reasoning
TCFT trains LLMs on temporal critique tasks to reduce post-cutoff knowledge leakage by 37-42 percentage points over prompting and standard SFT on Qwen models.
-
Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering
LLM agent progress depends on externalizing cognitive functions into memory, skills, protocols, and harness engineering that coordinates them reliably.
-
ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction
ZoFia is a zero-shot fake news detection framework that uses hierarchical entity salience retrieval followed by multi-LLM adversarial debate to improve robustness over single-model approaches.
-
Advancing Multi-Agent RAG Systems with Minimalist Reinforcement Learning
Mujica-MyGo decomposes multi-turn RAG interactions via multi-agent workflows and applies minimalist policy gradient optimization to improve performance on QA benchmarks while avoiding long-context problems.