DRIFTERSWriter / Director · 2024I GUESS I'LL ALWAYS HAVE YOUDirector · 2025FLESH AND BLOODWriter / Director · 2024MIDNIGHT FMWriter / Director · 2022DRIFTERSWriter / Director · 2024I GUESS I'LL ALWAYS HAVE YOUDirector · 2025FLESH AND BLOODWriter / Director · 2024MIDNIGHT FMWriter / Director · 2022DRIFTERSWriter / Director · 2024I GUESS I'LL ALWAYS HAVE YOUDirector · 2025FLESH AND BLOODWriter / Director · 2024MIDNIGHT FMWriter / Director · 2022
Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical NewsK. Gupta, D.Y. Liu, A. JoshiPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12863Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This pilot study explores explanation generation for satirical news articles from Australia and compares them alongside those from the USA. The paper demonstrates that framing explanations on unreliable narrator theory in open-weight models can be a feasible, cost-effective alternative to frontier models. Models identify satirical tropes more accurately on the American source than the Australian one. The work results in a novel dataset and an evaluation benchmark for satirical explanation.Language Models Have Interpretable Narrative MechanismsD.Y. Liu, A. Joshi, P. Dawson, A. PiperPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12989We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2 Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, while results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation and relation to narrative theory remain open questions.Scholarships & recognitionAwards supporting the MPhil at University of New South Wales (UNSW), in the UNSW NLP group.Research Training Program (RTP) Scholarship — UNSW, 2025–2027Domestic Research Training Grant (DRTG) — UNSW, 2026STEMM Champion — UNSW STEMM Champions Program, 2026Taste of Research Scholarship — UNSW, 2024AppointmentsTeaching and part-time engineering work alongside the MPhil.Casual Academic · UNSW · 2025–PresentI tutor across Computer Science and Engineering, and Arts and Media.Data Structures and AlgorithmsScreen Production IIFull Stack Software Engineer · Arludo · 2023–PresentI develop and maintain features for a digital education portal using React and AWS, part-time.Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical NewsK. Gupta, D.Y. Liu, A. JoshiPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12863Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This pilot study explores explanation generation for satirical news articles from Australia and compares them alongside those from the USA. The paper demonstrates that framing explanations on unreliable narrator theory in open-weight models can be a feasible, cost-effective alternative to frontier models. Models identify satirical tropes more accurately on the American source than the Australian one. The work results in a novel dataset and an evaluation benchmark for satirical explanation.Language Models Have Interpretable Narrative MechanismsD.Y. Liu, A. Joshi, P. Dawson, A. PiperPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12989We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2 Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, while results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation and relation to narrative theory remain open questions.Scholarships & recognitionAwards supporting the MPhil at University of New South Wales (UNSW), in the UNSW NLP group.Research Training Program (RTP) Scholarship — UNSW, 2025–2027Domestic Research Training Grant (DRTG) — UNSW, 2026STEMM Champion — UNSW STEMM Champions Program, 2026Taste of Research Scholarship — UNSW, 2024AppointmentsTeaching and part-time engineering work alongside the MPhil.Casual Academic · UNSW · 2025–PresentI tutor across Computer Science and Engineering, and Arts and Media.Data Structures and AlgorithmsScreen Production IIFull Stack Software Engineer · Arludo · 2023–PresentI develop and maintain features for a digital education portal using React and AWS, part-time.Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical NewsK. Gupta, D.Y. Liu, A. JoshiPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12863Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This pilot study explores explanation generation for satirical news articles from Australia and compares them alongside those from the USA. The paper demonstrates that framing explanations on unreliable narrator theory in open-weight models can be a feasible, cost-effective alternative to frontier models. Models identify satirical tropes more accurately on the American source than the Australian one. The work results in a novel dataset and an evaluation benchmark for satirical explanation.Language Models Have Interpretable Narrative MechanismsD.Y. Liu, A. Joshi, P. Dawson, A. PiperPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12989We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2 Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, while results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation and relation to narrative theory remain open questions.Scholarships & recognitionAwards supporting the MPhil at University of New South Wales (UNSW), in the UNSW NLP group.Research Training Program (RTP) Scholarship — UNSW, 2025–2027Domestic Research Training Grant (DRTG) — UNSW, 2026STEMM Champion — UNSW STEMM Champions Program, 2026Taste of Research Scholarship — UNSW, 2024AppointmentsTeaching and part-time engineering work alongside the MPhil.Casual Academic · UNSW · 2025–PresentI tutor across Computer Science and Engineering, and Arts and Media.Data Structures and AlgorithmsScreen Production IIFull Stack Software Engineer · Arludo · 2023–PresentI develop and maintain features for a digital education portal using React and AWS, part-time.Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical NewsK. Gupta, D.Y. Liu, A. JoshiPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12863Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This pilot study explores explanation generation for satirical news articles from Australia and compares them alongside those from the USA. The paper demonstrates that framing explanations on unreliable narrator theory in open-weight models can be a feasible, cost-effective alternative to frontier models. Models identify satirical tropes more accurately on the American source than the Australian one. The work results in a novel dataset and an evaluation benchmark for satirical explanation.Language Models Have Interpretable Narrative MechanismsD.Y. Liu, A. Joshi, P. Dawson, A. PiperPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12989We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2 Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, while results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation and relation to narrative theory remain open questions.Scholarships & recognitionAwards supporting the MPhil at University of New South Wales (UNSW), in the UNSW NLP group.Research Training Program (RTP) Scholarship — UNSW, 2025–2027Domestic Research Training Grant (DRTG) — UNSW, 2026STEMM Champion — UNSW STEMM Champions Program, 2026Taste of Research Scholarship — UNSW, 2024AppointmentsTeaching and part-time engineering work alongside the MPhil.Casual Academic · UNSW · 2025–PresentI tutor across Computer Science and Engineering, and Arts and Media.Data Structures and AlgorithmsScreen Production IIFull Stack Software Engineer · Arludo · 2023–PresentI develop and maintain features for a digital education portal using React and AWS, part-time.Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical NewsK. Gupta, D.Y. Liu, A. JoshiPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12863Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This pilot study explores explanation generation for satirical news articles from Australia and compares them alongside those from the USA. The paper demonstrates that framing explanations on unreliable narrator theory in open-weight models can be a feasible, cost-effective alternative to frontier models. Models identify satirical tropes more accurately on the American source than the Australian one. The work results in a novel dataset and an evaluation benchmark for satirical explanation.Language Models Have Interpretable Narrative MechanismsD.Y. Liu, A. Joshi, P. Dawson, A. PiperPreprint · 2026 · Narrative Theory & LLMsarXiv preprint arXiv:2610.12989We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2 Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, while results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation and relation to narrative theory remain open questions.Scholarships & recognitionAwards supporting the MPhil at University of New South Wales (UNSW), in the UNSW NLP group.Research Training Program (RTP) Scholarship — UNSW, 2025–2027Domestic Research Training Grant (DRTG) — UNSW, 2026STEMM Champion — UNSW STEMM Champions Program, 2026Taste of Research Scholarship — UNSW, 2024AppointmentsTeaching and part-time engineering work alongside the MPhil.Casual Academic · UNSW · 2025–PresentI tutor across Computer Science and Engineering, and Arts and Media.Data Structures and AlgorithmsScreen Production IIFull Stack Software Engineer · Arludo · 2023–PresentI develop and maintain features for a digital education portal using React and AWS, part-time.

DAVID YIYANG LIU

researcherfilmmaker