Houjun Liu

Houjun Liu, Lisa Einstein, John Yang et al.

LLM coding agents now generate code at an unprecedented scale, yet LLM-generated code introduces cybersecurity vulnerabilities into codebases without human involvement. Even when frontier models are explicitly asked to write secure production code with relevant weaknesses to avoid in context, we find that they still produce verifiable vulnerabilities on average 23% of the time across a corpus of 250 benign coding prompts. We introduce SecureForge, an automated pipeline that both audits security risks of frontier models and produces auditing-informed secure system prompts that reduce output security vulnerabilities while maintaining unit test performance. SecureForge first identifies benign prompts that produce statically detectable vulnerabilities, and then amplifies them into a large synthetic prompt corpus of diverse scenarios using a Markovian sampling technique to jointly maintain error rates and prompt diversity. This corpus is then used to iteratively optimize the system prompts to reduce output security vulnerabilities. On frontier models, SecureForge yields a statistically significant Pareto improvement in both unit test success and output security, with output vulnerabilities reduced by up to 48%. The resulting system prompts transfer zero-shot to in-the-wild coding agent prompts, without any exposure to real user prompt distributions during optimization.

Amelia F. Hardy, Houjun Liu, Allie Griffith et al.

Existing LLM red-teaming approaches prioritize high attack success rate, often resulting in high-perplexity prompts. This focus overlooks low-perplexity attacks that are more difficult to filter, more likely to arise during benign usage, and more impactful as negative downstream training examples. In response, we introduce ASTPrompter, a single-step optimization method that uses contrastive preference learning to train an attacker to maintain low perplexity while achieving a high attack success rate (ASR). ASTPrompter achieves an attack success rate 5.1 times higher on Llama-8.1B while using inputs that are 2.1 times more likely to occur according to the frozen LLM. Furthermore, our attack transfers to Mistral-7B, Qwen-7B, and TinyLlama in both black- and white-box settings. Lastly, by tuning a single hyperparameter in our method, we discover successful attack prefixes along an efficient frontier between ASR and perplexity, highlighting perplexity as a previously under-considered factor in red-teaming.

Houjun Liu, Brian MacWhinney

Language development researchers are interested in comparing the process of language learning across languages. Unfortunately, it has been difficult to construct a consistent quantitative framework for such comparisons. However, recent advances in AI (Artificial Intelligence) and ML (Machine Learning) are providing new methods for ASR (automatic speech recognition) and NLP (natural language processing) that can be brought to bear on this problem. Using the Batchalign2 program (Liu et al., 2023), we have been transcribing and linking data for the CHILDES database and have applied the UD (Universal Dependencies) framework to provide a consistent and comparable morphosyntactic analysis for 27 languages. These new resources open possibilities for deeper crosslinguistic study of language learning.

Houjun Liu

In this experiment, a model was devised, trained, and evaluated to automate psychotherapist/client text conversations through the use of state-of-the-art, Seq2Seq Transformer-based Natural Language Generation (NLG) systems. Through training the model upon a mix of the Cornell Movie Dialogue Corpus for language understanding and an open-source, anonymized, and public licensed psychotherapeutic dataset, the model achieved statistically significant performance in published, standardized qualitative benchmarks against human-written validation data - meeting or exceeding human-written responses' performance in 59.7% and 67.1% of the test set for two independent test methods respectively. Although the model cannot replace the work of psychotherapists entirely, its ability to synthesize human-appearing utterances for the majority of the test set serves as a promising step towards communizing and easing stigma at the psychotherapeutic point-of-care.

Houjun Liu, Shikhar Murty, Christopher D. Manning et al.

Current approaches for scaling inference-time compute in transformers rely on training them to emit explicit chain-of-thought tokens before producing an answer. While these methods are powerful, they are limited because they cannot be applied during pretraining and are limited to only serially-generated, natural-language verbalization to scale inference-time compute. In this work, we propose Thoughtbubbles, a transformer variant that natively performs parallel adaptive computation in latent space by learning to fork or delete residual streams. Thus, tokens that require a large amount of computation can form a "bubble" of cloned residuals in the middle of the network for additional thinking. Crucially, this behavior is learned during pretraining with only language modeling loss. Thoughtbubbles outperforms both standard decoder LMs as well as non-adaptive parallel computation approaches on OpenWebText and peS2o perplexity and in zero-shot evaluations such as HellaSwag and LAMBADA after pretraining across 150M to 772M parameter scales. The implicit nature of our method enables adaptive computation to be learned starting at pretraining time, paving the way to unify train and test-time behavior for reasoning models.

Houjun Liu, John Bauer, Christopher D. Manning · stanford

Originally, dropout was seen as a breakthrough regularization technique that reduced overfitting and improved performance in almost all applications of deep learning by reducing overfitting. Yet, single-epoch pretraining tasks common to modern LLMs yield minimal overfitting, leading to dropout not being used for large LLMs. Nevertheless, no thorough empirical investigation has been done on the role of dropout in LM pretraining. Through experiments in single-epoch pretraining of both masked (BERT) and autoregressive (Pythia 160M and 1.4B) LMs with varying levels of dropout, we find that downstream performance in language modeling, morpho-syntax (BLiMP), question answering (SQuAD), and natural-language inference (MNLI) improves when dropout is not applied during pretraining. We additionally find that the recently-introduced "early dropout" also degrades performance over applying no dropout at all. We further investigate the models' editability, and find that models trained without dropout are more successful in gradient-based model editing (MEND) and equivalent in representation-based model editing (ReFT). Therefore, we advocate to drop dropout during single-epoch pretraining.