Project summary

This project builds one instrument for working inside a language model, instead of the usual split where you need one tool to read internals, another to steer output, and another to edit weights. I fit a geometry to the model's own activations so that reading, steering and editing all happen in that same space. It works both in the residual stream and at the level of individual neurons.

The reason for building it that way is that reading and acting share one frame, so a measurement can be turned into an intervention and checked. If I claim some internal state means something, I can move that state and see whether the behaviour follows.

The problem this project would work on is a model that looks aligned in its chain of thought while representing something different underneath. Evaluation awareness is the first case: models that say, in their reasoning, that they suspect they are being tested. Reading that sentence is currently how the field notices it and in my current experiments I have been able to suppress it in COT, so its absence no longer tells you much. What I want to find out is whether I can suppress the awareness itself, not just the model saying it.

How will this funding be used?

Cloud GPU ~$7,000

Personal Expenses (3 mo ×$2,000) = $6,000

Subscriptions~ 800

Contingency~ $1,000

TOTAL~$15,000

Who is on your team? What's your track record on similar projects?

Just me.

I have previously co-authored a research paper on AI alignment(RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment).

What are the most likely causes and outcomes if this project fails?

I have run tests for my project on a smaller model (Qwen3.5-4B). I have been able to achieve the following:

1. Steering multiple regions of the model at once, a conserving rotation in its own geometry
2. Token readout, swap and steering: reading each layer's prediction, and swapping one token's direction inside the forward pass. The interaction is similar to the Jacobian lens, but the operator is different. Theirs exchanges the two directions, mine transfers the source component onto the target one and the swap direction comes from my own geometric equation rather than a fitted Jacobian.
3. Eval-awareness suppression: suppresing eval verbalised awareness seen in the thinking mode/COT( this removes the model saying it's being tested, not necessarily the awareness. That's what I want to work on.)

The most likely cause for the project to fail will be that the method I have built fails to find where the awareness sits. So, If the project fails on Eval awareness supression we will still have a novel model editing method tested on large open-source language models, and a research paper on it.

How much money have you raised in the last 12 months, and from where?

Currently, I have not raised money for this project.

Link to grantmaking.ai: https://app.grantmaking.ai/projects/f32f83db-8398-4cbe-86e0-f1df20467eb4