Aryaman Arora Acknowledgements
Learning a Generative Meta-Model of LLM Activations
Shaping capabilities with token-level data filtering
AutoMetrics: Approximate Human Judgements with Automatically
Generated Evaluators
Searching for Privacy Risks in LLM Agents via Simulation
Language Models May Verbatim Complete Text They Were Not
Explicitly Trained On
Soft production preferences emerge from a bottleneck on
memory
Self-refining diffusion samplers: Enabling parallelization via
parareal iterations
Attacking vision-language computer agents via pop-ups
From insights to actions: The impact of interpretability and
analysis research on NLP
Demystifying Verbatim Memorization in Large Language Models
Hydragen: High-Throughput LLM Inference with Shared Prefixes
Coding reliable LLM-based integrated task and knowledge agents
with GenieWorksheets
Representation fine-tuning on vision tasks
Sketch2Code: Evaluating vision-language models for interactive
web design prototyping
Improving access to untranscribed speech corpora using AI
Can LLMs generate novel research ideas? A large-scale human
study with 100+ NLP researchers
gzip predicts data-dependent scaling laws
A Universal Dependencies treebank for Gujarati
Connecting language technologies with rich, diverse data
sources covering thousands of languages
Improved neural protoform reconstruction via reflex
prediction
Design2Code: How far are we from automating front-end
engineering?
ColorSwap: A color and word order dataset for multimodal
evaluation
I am a Strange Dataset: Metalinguistic tests for language
models
Human raters cannot distinguish English translations from
original English texts
Verifying annotation agreement without multiple experts: A case
study with Gujarati SNACS
Semantic composition in visually grounded language models
Syntax-guided neural module distillation to probe
compositionality in sentence embeddings
Putting context in SNACS: A 5-Way classification of
adpositional pragmatic markers
Mischievous nominal constructions in Universal Dependencies
Results of the Second SIGMORPHON Shared Task on Multilingual
Grapheme-to-Phoneme Conversion
Draw *mir* a Sheep: A supersense-based analysis of German case
and adposition semantics