Pragmatic value alignment of LLMs to influence downstream behavoirs

Four-month project to fine-tune LLMs on public value surveys, evaluate behavior via a text-based agent simulator, and run a public consultation comparing model actions to respondents’ values.

Project Details

A 4 month research project in which I will:

fine-tune LLMs on public value survey data.
Build a text based agentic simulator to test for changes in behavior
Run a public consultation in which survey respondents review vignettes of models behavior and indicate which ones acted more in alignment with their values and "what they would have done".

It will be done solely by James Thompson (me) with supervision from two NZ academics. I have experience in applied AI (Tech lead on government agency internal AI system) as well as research experience working with training and evaluating LLMs. I feel this project is well suited to my skill sets.

There are two research questions that are being answered:

Can one use existing survey data to fine tune a model and get appreciable behavioral changes (tested through a novel text based behavioral evaluation)
Will end-users (i.e the public) prefer models from their own value group or others

The concrete impact is:

(Ideally) provide evidence for a pragmatic LLM alignment method that closer aligns its behavior to the end-user.

Of great importance as AI is being deployed in more autonomous decision making positions by actors who can't do large training runs and so small gains in alignment could have large downstream effect. Which is happening at some important government agencies in New Zealand (welfare office, police...)

(At the very least) add some more survey evidence to support a claim like "The public does/[not] like AI as it currently is operating"
Create a behavioral evaluation benchmark that can be used to compare various models real world decision making tendencies. This is intended to be made with a New Zealand context focus

Full funding request can be found here: https://raw.githubusercontent.com/1jamesthompson1/AIML589/main/docs/output/James-Thompson-LLM-NZ-Value-Alignment-Funding-Request.pdf

Theory of Impact

What I see going wrong currently

Current models are not going to be aligned to the end-user and for the public good in general. This is because they are developed without clear public accountability and incentives (i.e for-profit products), even if they were the pluralism of values is not accounted for when trying to aggregate values across all of humanity.

These models not being aligned to the end user does two things. To better understand them I would like to employ a visual of AI agents as being tools that will become the new arms and legs of humans (i.e the interface between our minds influencing the world).

Creates wide spread (albeit potentially small) mis-alignment between individuals intent and what actually happens in the real world. This is a first step towards gradual dis-empowerment as we are no longer influencing our futures as we thought we were.
Decreases trust between the greater public and the developers and deployers of AI systems. As I expect AI to be 'used anyway' this simply increases unrest and instability which is no-good-at-all.

How this addresses this

People

James Thompson

Team Member

Funding Details

Start Date: Jul 6, 2026
End Date: Nov 10, 2026
Expected Duration: 4 Months
Funding Raised to Date: -
Annual Budget: -
Monthly Burn Rate: -
Current Runway: -
Funding Goal: -
Funding Stage: seeking first grant
Fiscal Sponsor: -

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$5,000

Ideal

$6,500

How the money will be spent

Money will be spent on three things:

Compute costs 160 hours on a modern GPU rented in cloud (~H200 @ $3/hr): $480

Survey respondents 100 for 15 minutes: $850 (still waiting on quote as could be up to twice as much)

Living stipend to allow me to work full time without as many financial burdens $300 per week for 17 weeks: $5100

The minimum amount will fund the core research needs and some stipend.

The full amount will cover the full stipend and allow for either an increase in the survey size or models tested (which ever is deemed more impactful)

For more of a break down see this document:

https://raw.githubusercontent.com/1jamesthompson1/AIML589/main/docs/output/James-Thompson-LLM-NZ-Value-Alignment-Funding-Request.pdf#appendix.B

Discussion

No comments yet. Be the first to share your thoughts.