Towards Knowledgeable Foundation Models
@ ACL 2026 Workshop
San Diego, California, United States
Towards Knowledgeable Foundation Models
Foundation models have reshaped how AI systems acquire and utilize knowledge. Rather than relying solely on structured knowledge bases or curated documents, these models internalize vast amounts of world knowledge through large-scale pre-training, and can generate, retrieve, and reason over that knowledge at inference time.
Yet fundamental questions remain: Where does this knowledge come from? How much do foundation models know? Is their knowledge reliable and up-to-date? Can we control what they remember or forget? As models grow in scale and are deployed in multimodal, agentic, and retrieval-augmented settings, understanding and managing the knowledge lifecycle becomes increasingly critical.
This workshop examines the lifecycle of knowledge within foundation models across four key stages:
Emergence
How knowledge arises through pre-training and scaling
Injection
Augmenting models with external and retrieved knowledge
Updating
Editing, correcting, and erasing knowledge in models
Probing & Generation
Evaluating, extracting, and generating knowledge
This is the 4th KnowFM workshop. Previous editions: KnowFM@ACL2025, KnowFM@AAAI2025, and KnowLM@ACL2024.
Call for Papers
Foundation models have demonstrated remarkable capabilities in storing and utilizing knowledge acquired during pre-training. Yet as these models scale and are deployed in increasingly complex settings, critical challenges remain: How can we reliably assess what a model knows? How do we reconcile conflicting knowledge from parametric memory and retrieved context? How can we keep model knowledge up-to-date without compromising reasoning abilities? And how do we extend these capabilities beyond text to multimodal and agentic settings?
This workshop brings together researchers working on different stages and aspects of the knowledge lifecycle, from structured and unstructured knowledge sources to knowledge acquired and synthesized by models themselves, to discuss how knowledge should be represented, acquired, verified, and applied in the era of foundation models.
Submission Topics
We welcome submissions on all topics related to knowledgeable foundation models, including:
- Analysis of knowledge within foundation models: how much they know, where that knowledge comes from, and how it is represented
- Enhancing models with existing knowledge sources (knowledge graphs, domain-specific databases, manuals, rules, etc.) during training or inference
- Analyzing and improving RAG (retrieval-augmented generation) systems
- Updating, editing, and erasing knowledge in foundation models
- Knowledge extraction, generation, and distillation using foundation models
- Synthetic data quality and reliability for knowledge-intensive tasks
- Evaluation of knowledge utilization (faithfulness, truthfulness, attribution) by foundation models
- Identification and mitigation of hallucinations and factual errors
- Knowledge conflicts: resolving inconsistencies between parametric memory and retrieved or long-context information
- Knowledge in multimodal foundation models: visual knowledge, cross-modal grounding, and multimodal RAG
- Knowledge-intensive agents: search agents, tool-augmented agents, and agentic RAG systems
- Grounding and knowledge acquisition in multi-step reasoning and planning agents
Paper Awards
We will also announce a Best Paper Award and an Outstanding Paper Award at our workshop.
Submission Instructions
We welcome two types of papers: regular workshop papers and non-archival submissions. Only regular workshop papers will be included in the workshop proceedings. Review process will be double-blind. All submissions should be in PDF format following the ACL template (8 pages for main text) and made through OpenReview submission portal (https://openreview.net/group?id=aclweb.org/ACL/2026/Workshop/KnowFM)
Important Dates
All deadlines are 23:59pm UTC-12h ("Anywhere on Earth").
Submission deadline is set after the ACL 2026 notification of acceptance.
Tentative Speakers
Luna Dong
Meta
Weijia Shi
University of Washington
Lianhui Qin
UC San Diego
Jiaxin Pei
Stanford University
Parisa Kordjamshidi
Michigan State University
Schedule
July 3, 2026 · KnowFM @ ACL 2026
All times shown in Pacific Time (PT)
- When Do LLMs Admit Their Mistakes? Understanding the Role of Model Belief in Retraction
- Emergent Unfaithfulness: How Alignment Training Causes LMs to Silently Override Task Faithfulness
- Rule vs. Consequence: Dissociable Internal Representations of Moral Reasoning in LLMs
- BaseCal: Unsupervised Confidence Calibration via Base Model Signals
- FaithLens: Detecting and Explaining Faithfulness Hallucination
- ScienceMeter: Tracking Scientific Knowledge Updates in Language Models
- Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning
Imagine a personal assistant that, with user permission, persistently remembers moments from daily life—answering questions like “When and where did I see this lady?” or offering personalized suggestions like “You might enjoy The Little Prince—it relates to the statue you liked in Lyon.” Realizing this vision requires overcoming major challenges: capturing visual memories under hardware constraints (e.g., memory, battery, thermal limits, bandwidth), extracting meaningful personalization signals from noisy, task-agnostic visual histories, and supporting real-time question answering and recommendations under tight latency requirements.
In this talk, we present our early work toward this goal. Pensieve, our memory-based QA system, improves accuracy by 11% over state-of-the-art multimodal RAG baselines. VisualLens infers user interests from casual photos, outperforming leading recommendation systems by 5–10%. We also share initial results on efficient, event-triggered memory capture and compression. Our work points to a broad landscape of research opportunities in building richer, more context-aware personal assistants capable of learning from and reasoning over users’ visual experiences.
Bio. Xin Luna Dong is a Principal Scientist at Meta Wearables AI, where she leads the Agentic AI efforts for building trustworthy and personalized assistants on wearable devices. Previously, she spent over a decade advancing knowledge graph technology, including the Amazon Product Graph and the Google Knowledge Graph. She is co-author of Machine Knowledge: Creation and Curation of Comprehensive Knowledge Bases and Big Data Integration. She was named an ACM Fellow and an IEEE Fellow for “significant contributions to knowledge graph construction and data integration”, awarded the VLDB Women in Database Research Award and VLDB Early Career Research Contribution Award, and invited as an ACM Distinguished Speaker. She was a member of the VLDB endowment, a PC co-chair for ICDM’2027, KDD’2022 ADS track, WSDM’2022, VLDB’2021, and Sigmod’2018.
Organizers
Organizing Committee
Canyu Chen
Northwestern University
Yuji Zhang
University of Illinois Urbana-Champaign
Zoey Sha Li
Amazon
Zihan Wang
Northwestern University
Qineng Wang
Northwestern University
Jinyan Su
Cornell University
Priyanka Kargupta
University of Illinois Urbana-Champaign
Sara Vera Marjanović
University of Copenhagen
Jeff Z. Pan
University of Edinburgh
Manling Li
Northwestern University
Advising Committee
Heng Ji
University of Illinois Urbana-Champaign
Mohit Bansal
University of North Carolina at Chapel Hill
Isabelle Augenstein
University of Copenhagen
