Scaling Trust by Quantifying Risk in Frontier Generative Foundation Models

Generative foundation models (GenFMs), such as large language models, are reshaping our society. Despite their potential, their deployment in safety-critical domains like healthcare, law, and finance remains limited by hallucinations, where models produce factually incorrect but convincing outputs. This problem not only introduces tangible risks but also erodes public trust in AI systems, preventing their wider adoption. Existing solutions, such as retrieval-augmented generation, offer only partial remedies and often introduce new complexities. More fundamentally, uncertainty quantification (UQ) provides a model-internal mechanism to estimate the reliability of generated content. However, there are key challenges in making UQ for GenFMs efficient, reliable, and generalisable. Solving these challenges is critical to unlocking the safe and responsible use of AI.

The STQR project proposes to scale the trust in GenFMs by quantifying their risks through the research of a novel generative UQ (GenUQ) framework with efficiency, reliability and expanded scope. GenUQ will incorporate uncertainty estimation into the generation process to make it more efficient, develop calibration techniques that align uncertainty scores with the truthfulness of outputs, and build modular, probabilistic models capable of generalising across diverse user domains. By advancing methods such as conformal prediction, probabilistic modelling, preference optimisation and efficient inference, the project aims to create fine-grained, reliable, and explainable uncertainty measures that scale with the growing capabilities of AI models.

In healthcare and other safety-critical fields, STQR will flag unreliable outputs and reduce risks to patients and users. For policymakers and regulators, the project will provide tools to audit and certify AI systems more effectively, supporting national and international AI safety initiatives. For the research community, it will open new directions in scalable risk assessment. The project’s outputs will also contribute to making GenUQ a practical feature for the next generation of AI applications, boosting greater public trust in the AI systems.

As part of this project, research activities can include (non-exhaustive):

· The design, deployment and evaluation of scalable uncertainty quantification that generalize to long-horizon tasks;

· The application of scalable uncertainty quantification in safety-critical tasks, such as healthcare and education;

· The design, deployment and evaluation of de-centralized uncertainty quantification that has the potential to evaluate GenFMs in different domains.

The PhD student will receive tuition fees and a London stipend at UKRI rates (currently in 2024/25 of £21,237 per year, to be confirmed for 2025/26) annually during the PhD period, which can span for 3 years.

For more information about the project, please contact Ziquan Liu (ziquan.liu@qmul.ac.uk).

Supervisor

Dr Ziquan Liu (he/him) – ziquan.liu@qmul.ac.uk

Webpage: https://sites.google.com/view/ziquanliu

Centre for Multimodal AI: https://www.seresearch.qmul.ac.uk/cmai/

Google Scholar: https://scholar.google.com/citations?user=x28OqBkAAAAJ&hl=en

How to apply

Queen Mary is interested in developing the next generation of outstanding researchers and decided to invest in specific research areas. For further information about potential PhD projects and supervi-sors please see the list of the projects at the end of this page.

Applicants should work with their prospective supervisor and submit their application following the in-structions at: http://eecs.qmul.ac.uk/phd/how-to-apply/

The application should include the following:

· CV (max 2 pages)

· Cover letter (max 4,500 characters) stating clearly in the first page whether you are eligible for a scholarship as a UK resident (https://epsrc.ukri.org/skills/students/guidance-on-epsrc-stu-dentships/eligibility)

· Research proposal (max 500 words)

· 2 References

· Certificate of English Language (for students whose first language is not English)

· Other Certificates

Please note that to qualify as a home student for the purpose of the scholarships, a student must have no restrictions on how long they can stay in the UK and have been ordinarily resident in the UK for at least 3 years prior to the start of the studentship. For more information please see: (https://epsrc.ukri.org/skills/students/guidance-on-epsrc-studentships/eligibility)

Application Deadline

The deadline for applications is the 24th November 2025.

For general enquiries contact Mrs Melissa Yeo at m.yeo@qmul.ac.uk (administrative enquiries) or Dr Arkaitz Zubiaga at a.zubiaga@qmul.ac.uk (academic enquiries) with the subject “EECS 2025 PhD schol-arships enquiry”. For specific enquiries contact Dr Ziquan Liu at ziquan.liu@qmul.ac.uk