# Gavin Nguyen

> Personal site on payments technology and reliable language models. Paper pages state the problem first, then a short hint of the approach.

## Research papers

- [Twin Worlds: Equivariance-Based Abstention for Evidence-Grounded Reasoning](https://itsgavin.me/research/twin-worlds): A model can be confident and still be wrong because its memory reflects an older world.
- [Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models](https://itsgavin.me/research/hallucinate-less-by-thinking-more): When a question feels familiar, a model jumps to the first reading that fits. I bring the other readings back before it commits.
- [Bootstrap Wayfinding Questions to Elicit Emotion Shift Reasoning with Large Language Models](https://itsgavin.me/research/bootstrap-wayfinding-questions): When emotion changes in a conversation, the question is what caused the change, not what each sentence felt. I start from that change and reason towards its cause.
- [Emotion Flip Reasoning via Stacked Instruction Finetuning of LLMs](https://itsgavin.me/research/emotion-flip-reasoning): Naming the trigger in one prediction hides the judgments that identify it. I train those judgments in steps.

## Site

- [Home](https://itsgavin.me/): Introduction.
- [Work](https://itsgavin.me/work): Payments company, from infrastructure to scale.
- [Research](https://itsgavin.me/research): Reliable language models, hallucination, abstention and post-training behaviour.
- [Writing](https://itsgavin.me/writing): Notes on payments, AI and technology.
- [Google Scholar](https://scholar.google.com/citations?user=-eEmuqgAAAAJ): Full publication list.

## Writing

- [When should an AI system say "I don't know"?](https://itsgavin.me/writing/when-should-an-ai-system-say-i-dont-know): Reliability sometimes means knowing when not to answer.
- [What payment systems taught me about reliability](https://itsgavin.me/writing/what-payment-systems-taught-me-about-reliability): Some lessons about designing systems where failures are unavoidable.
- [Why useful AI is not necessarily reliable AI](https://itsgavin.me/writing/why-useful-ai-is-not-necessarily-reliable-ai): Training a model to be more helpful can also make it less careful about when that help is warranted.
- [Why local payment rails matter](https://itsgavin.me/writing/why-local-payment-rails-matter): Cross-border products still have to clear locally. The rail is not an implementation detail.
- [What changes when you move from engineer to CTO](https://itsgavin.me/writing/what-changes-when-you-move-from-engineer-to-cto): The work shifts from making the system correct to making the organisation able to keep it correct.