Modeling and Predicting Multi-Turn Answer Instability in Large Language Models

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	He, Jiahang, Ramachandran, Rishi, Ramachandran, Neel, Katakam, Aryan, Zhu, Kevin, Dev, Sunishchal, Panda, Ashwinee, Shrivastava, Aryan
Format:	Preprint
Veröffentlicht:	2025
Schlagworte:	Computation and Language
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Ähnliche Einträge

Linearly Decoding Refused Knowledge in Aligned Language Models
von: Shrivastava, Aryan, et al.
Veröffentlicht: (2025)

DICE: A Framework for Dimensional and Contextual Evaluation of Language Models
von: Shrivastava, Aryan, et al.
Veröffentlicht: (2025)

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models
von: Jain, Neel, et al.
Veröffentlicht: (2024)

Measuring Free-Form Decision-Making Inconsistency of Language Models in Military Crisis Simulations
von: Shrivastava, Aryan, et al.
Veröffentlicht: (2024)

Advancing Reasoning in Large Language Models: Promising Methods and Approaches
von: Patil, Avinash, et al.
Veröffentlicht: (2025)

Resource-Aware Arabic LLM Creation: Model Adaptation, Integration, and Multi-Domain Testing
von: Aryan, Prakash
Veröffentlicht: (2024)

SALT: Steering Activations towards Leakage-free Thinking in Chain of Thought
von: Batra, Shourya, et al.
Veröffentlicht: (2025)

Structure First, Reason Next: Enhancing a Large Language Model using Knowledge Graph for Numerical Reasoning in Financial Documents
von: Mishra, Aryan, et al.
Veröffentlicht: (2026)

Causal Reflection with Language Models
von: Aryan, Abi, et al.
Veröffentlicht: (2025)

A Comparative Study of Translation Bias and Accuracy in Multilingual Large Language Models for Cross-Language Claim Verification
von: Singhal, Aryan, et al.
Veröffentlicht: (2024)

Adapting Biomedical Abstracts into Plain language using Large Language Models
von: Gangavarapu, Haritha, et al.
Veröffentlicht: (2025)

AbsenceBench: Language Models Can't Tell What's Missing
von: Fu, Harvey Yiyun, et al.
Veröffentlicht: (2025)

Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMs
von: Afonin, Nikita, et al.
Veröffentlicht: (2025)

English Please: Evaluating Machine Translation with Large Language Models for Multilingual Bug Reports
von: Patil, Avinash, et al.
Veröffentlicht: (2025)

Private Fine-tuning of Large Language Models with Zeroth-order Optimization
von: Tang, Xinyu, et al.
Veröffentlicht: (2024)

Large Language Models aren't all that you need
von: Holla, Kiran Voderhobli, et al.
Veröffentlicht: (2024)

Signal or Noise? Evaluating Large Language Models in Resume Screening Across Contextual Variations and Human Expert Benchmarks
von: Varshney, Aryan, et al.
Veröffentlicht: (2025)

AI Based Font Pair Suggestion Modelling For Graphic Design
von: Singh, Aryan, et al.
Veröffentlicht: (2025)

Know Thyself? On the Incapability and Implications of AI Self-Recognition
von: Bai, Xiaoyan, et al.
Veröffentlicht: (2025)

Multi-Token Prediction via Self-Distillation
von: Kirchenbauer, John, et al.
Veröffentlicht: (2026)

Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Models
von: Lin, Zizhuo, et al.
Veröffentlicht: (2026)

Grounded Concreteness: Human-Like Concreteness Sensitivity in Vision-Language Models
von: Roy, Aryan, et al.
Veröffentlicht: (2026)

Investigating Spatial Attention Bias in Vision-Language Models
von: Chaudhary, Aryan, et al.
Veröffentlicht: (2025)

Visualizing and Benchmarking LLM Factual Hallucination Tendencies via Internal State Analysis and Clustering
von: Mao, Nathan, et al.
Veröffentlicht: (2026)

TurnBench-MS: A Benchmark for Evaluating Multi-Turn, Multi-Step Reasoning in Large Language Models
von: Zhang, Yiran, et al.
Veröffentlicht: (2025)

Is Training Data Quality or Quantity More Impactful to Small Language Model Performance?
von: Sajith, Aryan, et al.
Veröffentlicht: (2024)

Alignment-Constrained Dynamic Pruning for LLMs: Identifying and Preserving Alignment-Critical Circuits
von: Patel, Dev, et al.
Veröffentlicht: (2025)

LLMs as Debate Partners: Utilizing Genetic Algorithms and Adversarial Search for Adaptive Arguments
von: Aryan, Prakash
Veröffentlicht: (2024)

Illuminate: A novel approach for depression detection with explainable analysis and proactive therapy using prompt engineering
von: Agrawal, Aryan
Veröffentlicht: (2024)

TRUTH DECAY: Quantifying Multi-Turn Sycophancy in Language Models
von: Liu, Joshua, et al.
Veröffentlicht: (2025)

COMPASS: Context-Modulated PID Attention Steering System for Hallucination Mitigation
von: Sahay, Kenji, et al.
Veröffentlicht: (2025)

DebateBench: A Challenging Long Context Reasoning Benchmark For Large Language Models
von: Tiwari, Utkarsh, et al.
Veröffentlicht: (2025)

The Text Uncanny Valley: Non-Monotonic Performance Degradation in LLM Information Retrieval
von: Tong, Zekai, et al.
Veröffentlicht: (2026)

LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation
von: Zhang, Juzheng, et al.
Veröffentlicht: (2025)

Amortized Latent Steering: Low-Cost Alternative to Test-Time Optimization
von: Egbuna, Nathan, et al.
Veröffentlicht: (2025)

The Chameleon Nature of LLMs: Quantifying Multi-Turn Stance Instability in Search-Enabled Language Models
von: Ratnakar, Shivam, et al.
Veröffentlicht: (2025)

Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory
von: Chhikara, Prateek, et al.
Veröffentlicht: (2025)

Teach LLMs to Phish: Stealing Private Information from Language Models
von: Panda, Ashwinee, et al.
Veröffentlicht: (2024)

Question: How do Large Language Models perform on the Question Answering tasks? Answer:
von: Fischer, Kevin, et al.
Veröffentlicht: (2024)

Align to Structure: Aligning Large Language Models with Structural Information
von: Kim, Zae Myung, et al.
Veröffentlicht: (2025)

if (filemtime($filename) + $maxlifetime < time()) { unlink($filename); $count++; } } return $count; } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ protected function saveSession($sessId, $data): bool { $sessFile = $this->path . '/sess_' . $sessId; if ($handle = fopen($sessFile, 'w')) { $return = false; // Lock the file for exclusive access to avoid issues with multiple // processes writing session simultaneously: if (flock($handle, LOCK_EX)) { $return = fwrite($handle, $data); // Make sure that there's no trailing data by truncating the file to // the correct length: ftruncate($handle, strlen($data)); fflush($handle); flock($handle, LOCK_UN); } fclose($handle); if ($return !== false) { return true; } } // If we got this far, something went wrong with the file output... // It is tempting to throw an exception here, but this code is called // outside of the context of exception handling, so all we can do is // echo a message.

// Anecdotal testing Today and Yesterday seems to indicate destroy() // is called by the garbage collector and everything is good. // Something to keep in mind though. return 0; } /** * Write function that is called when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ public function write($sessId, $data): bool { if ($this->writesDisabled) { return true; } return $this->saveSession($sessId, $data); } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ abstract protected function saveSession($sessId, $data): bool; }

* * @return void */ public function writeClose() { // The assumption is that we're using PHP's ext/session. // session_write_close() will actually overwrite $_SESSION with an // empty array on completion -- which leads to a mismatch between what // is in the storage object and $_SESSION. To get around this, we // temporarily reset $_SESSION to an array, and then re-link it to // the storage object. // // Additionally, while you _can_ write to $_SESSION following a // session_write_close() operation, no changes made to it will be // flushed to the session handler. As such, we now mark the storage // object isImmutable. $storage = $this->getStorage(); if (! $storage->isImmutable()) { $_SESSION = $storage->toArray(true); session_write_close(); $storage->fromArray($_SESSION); $storage->markImmutable(); } } /** * Attempt to set the session name * * If the session has already been started, or if the name provided fails * validation, an exception will be raised. * * @param string $name * @return SessionManager * @throws Exception\InvalidArgumentException */ public function setName($name) { if ($this->sessionExists()) { throw new Exception\InvalidArgumentException( 'Cannot set session name after a session has already started'

Environment & details:

GET Data empty

POST Data empty

Files empty

Key	Value
language	"de"
ui	"standard"
VUFIND_SESSION	"bb900e19343dae92211ea248981b777e"

Session

Key	Value
__Laminas	array:2 [`"_REQUEST_ACCESS_TIME" => 1790870931.9661 "_VALID" => array:1 [ "Laminas\Session\Validator\Id" => "bb900e19343dae92211ea248981b777e" ]`]
SessionState	Laminas\Stdlib\ArrayObject {#938}
FlashMessenger	Laminas\Stdlib\ArrayObject {#1433}
SessionHelper	Laminas\Stdlib\ArrayObject {#1759}

Server/Request Data

Key	Value
REDIRECT_VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
REDIRECT_HTTPS	"on"
REDIRECT_SSL_TLS_SNI	"info.umar.mx"
REDIRECT_STATUS	"200"
VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
HTTPS	"on"
SSL_TLS_SNI	"info.umar.mx"
HTTP_ACCEPT	"/"
HTTP_USER_AGENT	"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)"
HTTP_COOKIE	"language=de; ui=standard; VUFIND_SESSION=bb900e19343dae92211ea248981b777e"
HTTP_ACCEPT_ENCODING	"gzip, br, zstd, deflate"
HTTP_HOST	"info.umar.mx"
PATH	"/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/snap/bin"
SERVER_SIGNATURE	"<address>Apache/2.4.66 (Ubuntu) Server at info.umar.mx Port 443</address>\n"
SERVER_SOFTWARE	"Apache/2.4.66 (Ubuntu)"
SERVER_NAME	"info.umar.mx"
SERVER_ADDR	"200.23.222.19"
SERVER_PORT	"443"
REMOTE_ADDR	"216.73.216.147"
DOCUMENT_ROOT	"/usr/local/vufind/public"
REQUEST_SCHEME	"https"
CONTEXT_PREFIX	""
CONTEXT_DOCUMENT_ROOT	"/usr/local/vufind/public"
SERVER_ADMIN	"admin@localhost"
SCRIPT_FILENAME	"/usr/local/vufind/public/index.php"
REMOTE_PORT	"63453"
REDIRECT_URL	""" /Record/\n arxiv_https___arxiv_org_abs_2511_10688/Similar """
GATEWAY_INTERFACE	"CGI/1.1"
SERVER_PROTOCOL	"HTTP/1.1"
REQUEST_METHOD	"GET"
QUERY_STRING	""
REQUEST_URI	"/Record/%0Aarxiv_https___arxiv_org_abs_2511_10688/Similar"
SCRIPT_NAME	"/index.php"
PHP_SELF	"/index.php"
REQUEST_TIME_FLOAT	1790870931.9611
REQUEST_TIME	1790870931

Environment Variables empty

Registered Handlers

0. Whoops\Handler\PrettyPageHandler