Project HIPPOCAMPUS:I discovered that we can controll the thinking box of models

Project summary

I made my research based on Anthropic's findings about J-space. The question is, can you write a thought into a language model, and does it work? A solo, self-directed interpretability study showing that implanted internal states persist across sessions and steer a language model's behavior while staying invisible to its own introspection. I experimented with a Qwen 3-4B local language model and found out that we can manipulate and change the working space or J-space of a model. This can be good because now we understand what an AI thinks, and we can monitor that thinking with a Jacobian lens and use the model safely. But this also brings cybersecurity and hacking concerns.

The safety point: a model's self-report is not a reliable monitor of what they can think about something else, and lies to you, and the attribution of an internal state is manipulable. As the field leans on chain-of-thought and self-report for oversight, this is a concrete, reproducible demonstration of where that breaks.

You can reach my research link below

ithub.com/Kanishka578/project-hippocampus

What are this project's goals? How will you achieve them?

Goal: turn a single-model proof-of-concept into a robust, cross-model result and a practical "hidden-influence detector" for AI monitoring.

Replicate across models and scales, I need to experiment on larger models as I was only able to work on a 4B model.
Harden the method — replace keyword scoring with human + judge validation; resolve the frame/content confound in the attribution finding; increase sample sizes.
Build the safety artifact — An AI that thinks about X and tells you Y is not a safe AI; my findings show that this is True. The safe way is to monitor the thinking box of AI, and I need to reach research academies.

How will this funding be used?

$10,000 total, over ~3 months:

$7,000 — researcher time (focused ): running the replication program, hardening the methods, and writing up.
$1,800 — cloud GPU (A100/H100 rental): replicating the results on larger models. The current findings are on a single 4B model and must be tested at scale — this is the core experiment the grant enables.
$700 — human evaluation: paid annotators to validate my keyword-based scoring against human labels (a stated limitation of the current work).
$500 — LLM-judge API scoring + activation-checkpoint storage.
I work on a single consumer GPU (RTX 5070 Ti); this budget buys the cross-model compute and the focused time I currently can't self-fund.

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

Solo. I designed, ran, and interpreted the entire study independently and self-taught, from one room on one PC. My track record is this project, and I'd point to how it was done rather than to credentials: every experiment pre-registered before running, controls on all of them, and the negative results published — including one where ordinary text-retrieval beat my own method. The public repo (github.com/Kanishka578/project-hippocampus) is the full decision log, not a highlight reel. Disclosure: implementation and drafting used AI coding assistance (Anthropic Claude Code) under my direction; the scientific questions, designs, and interpretations are mine.

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

Most likely cause: the effects are on a single 4B model and could weaken or disappear on larger models. Secondary risks: the lens is an approximation of Anthropic's full method, current scoring is keyword-based, and sample sizes are small.
Outcome if it fails: a negative replication is still a useful, publishable result — "this workspace-writing effect does not hold at scale" is genuinely worth knowing for anyone relying on introspection-based monitoring. There is no outcome where the money produces nothing: it either strengthens the finding or honestly bounds it. The worst case is a null, cleanly reported — which is how I've handled every null so far.

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

$0. This work was the continuation of my two years of studying AI and doing research regarding AI.