# Limits to Predicting Online Speech Using Large Language Models

arXiv: https://arxiv.org/abs/2407.12850 · HTML: https://mlsanity.com/paper/2407.12850/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 54 references that are in the map.
- First posted on arXiv 8 Jul 2024.

## Builds on (most cited first, 23 of 23)

- [Ethical and social risks of harm from Language Models](https://mlsanity.com/paper/2112.04359/) (2021)
- [LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale](https://mlsanity.com/paper/2208.07339/) (2022)
- [Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation](https://mlsanity.com/paper/2302.09664/) (2023)
- [The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only](https://mlsanity.com/paper/2306.01116/) (2023)
- [The Falcon Series of Open Language Models](https://mlsanity.com/paper/2311.16867/) (2023)
- [Analyzing Leakage of Personally Identifiable Information in Language Models](https://mlsanity.com/paper/2302.00539/) (2023)
- [Demographic Dialectal Variation in Social Media: A Case Study of African-American English](https://mlsanity.com/paper/1608.08868/) (2016)
- [What Does it Mean for a Language Model to Preserve Privacy?](https://mlsanity.com/paper/2202.05520/) (2022)
- [Beyond Memorization: Violating Privacy Via Inference with Large Language Models](https://mlsanity.com/paper/2310.07298/) (2023)
- [DeepInf: Social Influence Prediction with Deep Learning](https://mlsanity.com/paper/1807.05560/) (2018)
- [Uncertainty Quantification with Pre-trained Language Models: A Large-Scale Empirical Analysis](https://mlsanity.com/paper/2210.04714/) (2022)
- [A Mathematical Exploration of Why Language Models Help Solve Downstream Tasks](https://mlsanity.com/paper/2010.03648/) (2020)
- [Theories of "Gender" in NLP Bias Research](https://mlsanity.com/paper/2205.02526/) (2022)
- [Scalable and Generalizable Social Bot Detection through Data Selection](https://mlsanity.com/paper/1911.09179/) (2019)
- [Characterizing Manipulation from AI Systems](https://mlsanity.com/paper/2303.09387/) (2023)
- [Detection of Novel Social Bots by Ensembles of Specialized Classifiers](https://mlsanity.com/paper/2006.06867/) (2020)
- [Retweet-BERT: Political Leaning Detection Using Language Features and Information Diffusion on Social Networks](https://mlsanity.com/paper/2207.08349/) (2022)
- [TweetBERT: A Pretrained Language Representation Model for Twitter Text Analysis](https://mlsanity.com/paper/2010.11091/) (2020)
- [Predicting Successful Memes using Network and Community Structure](https://mlsanity.com/paper/1403.6199/) (2014)
- [Social-LLM: Modeling User Behavior at Scale using Language Models and Social Network Data](https://mlsanity.com/paper/2401.00893/) (2023)
- [Using Embeddings for Causal Estimation of Peer Influence in Social Networks](https://mlsanity.com/paper/2205.08033/) (2022)
- [Information flow reveals prediction limits in online social activity](https://mlsanity.com/paper/1708.04575/) (2017)
- [SMILE: Evaluation and Domain Adaptation for Social Media Language Understanding](https://mlsanity.com/paper/2307.00135/) (2023)

## Abstract (arXiv)

Our paper studies the predictability of online speech -- that is, how well language models learn to model the distribution of user generated content on X (previously Twitter). We define predictability as a measure of the model's uncertainty, i.e. its negative log-likelihood. As the basis of our study, we collect 10M tweets for ``tweet-tuning'' base models and a further 6.25M posts from more than five thousand X (previously Twitter) users and their peers. In our study involving more than 5000 subjects, we find that predicting posts of individual users remains surprisingly hard. Moreover, it matters greatly what context is used: models using the users' own history significantly outperform models using posts from their social circle. We validate these results across four large language models ranging in size from 1.5 billion to 70 billion parameters. Moreover, our results replicate if instead of prompting the model with additional context, we finetune on it. We follow up with a detailed investigation on what is learned in-context and a demographic analysis. Up to 20% of what is learned in-context is the use of @-mentions and hashtags. Our main results hold across the demographic groups we studied.