Suchir Salhan

Name:	Suchir Salhan
Affiliation:	University of Cambridge
E-mail:	(only provided to users who are logged into talks.cam)
Last login:	18 Mar 2026, 10:32 a.m.

Public lists managed by Suchir Salhan

• NLIP Seminar Series

Talks given by Suchir Salhan

Obviously this only lists talks that are listed through talks.cam. Furthermore, this facility only works if the speaker's e-mail was specified in a talk. Most talks have not done this.

• NLIP 2024 Social: Meet New PhD Students

Talks organised by Suchir Salhan

This list is based on what was entered into the 'organiser' field in a talk. It may not mean that Suchir Salhan actually organised the talk, they may have been responsible only for entering the talk into the talks.cam system.

• Title to be confirmed
• Understanding the Interplay between LLMs' Utilisation of Parametric and Contextual Knowledge
• CodeScaler: Scaling Code LLM Training and Test-Time Inference via Execution-Free Reward Models
• AI Metrics: Theoretical Foundations, Design, and Selection of Evaluation Metrics Based on Ground Truth
• Generalised measures of predictive uncertainty in online language processing
• What Happens When They're Smarter Than Us?
• Large Language Models' Complicit Responses to Illicit Instructions across Socio-Legal Contexts
• The emerging causal paradigm in interpretability
• Memorization as a Feature, Not a Bug
• Competence, Pain and the Geometry of Intelligence: Three Perspectives on Learning and Representation
• Building and Understanding Human-scale Language Models
• Title to be confirmed
• A computational model of language production
• With great power... My journey in Responsible AI
• Extrapolating model performance across training horizons
• Title to be confirmed
• Title to be confirmed
• Evaluation with LLMs - Theoretical and Practical insights
• Making and breaking tokenizers
• NLIP 2025 Social: Meet New PhD Students
• LLMs, Implicit Bayesian inference and compositional Generalization
• MultiBLiMP: A Multilingual Benchmark of Linguistic Minimal Pairs
• Title to be confirmed
• When is Multilinguality a Curse? Language Modeling for 350 Languages
• Title to be confirmed
• Unveiling the Secret Sauce: A Causal Look at Data Memorisation and Tokenisation in Language Models
• Robust Alignment of Large Language Models
• Measuring Political Bias in Large Language Models
• Research Progress in Mechanistic Interpretability
• Asymmetry in Supposedly Equivalent Facts: Pre-training Bias in Large Language Models
• Title to be confirmed
• Title to be confirmed
• Formal syntactic theory in the current NLP landscape
• Natural Language meets Control Theory
• Typological Diversity in NLP: What, Why and a Way Forward
• Title to be confirmed
• Assessing language-specific capabilities of LLMs: Lessons from Swedish NLP
• Metrized Deep Learning: Fast & Scalable Training
• LLMs as supersloppers and other metaphors
• Preference Alignment, with Reference Mismatch, and without Reference Models
• Analysing Memorisation in Classification and Translation through Localisation and Cartography
• The Past, Present and Future of Tokenization
• Title to be confirmed
• Scansion-based Lyric Generation
• Linguistics in the Age of Large Language Models.
• 10 Slides on Human Feedback
• Adaptive Tokenization and Memory in Foundation Models
• Language Modelling with Phonemes
• Truth conditions at scale, and beyond
• NLIP 2024 Social: Meet New PhD Students