Research Before Claims

Research practice

What the method needs to distinguish

If two AI systems develop different abilities, the difference could come from their tasks, their starting conditions or the way the experiment measures capability. NT-CIE is intended to isolate the role of long-term needs. Whether it does so reliably is still under review.

Early experiments gave us positive and negative results. Comparisons with stronger alternative methods exposed weaknesses, so we narrowed or removed some of the proposed mechanisms.

We use pilot results to find weaknesses in the experiment. We also need comparisons with alternative methods, records of negative results and checks that a run can be reproduced. Pilot results do not confirm Need-Shaped Personal AI.

Public update / 23 September 2026

What Paper 01 now asks

Paper 01 asks whether changing needs can influence which abilities an AI develops and retains. NT-CIE is our proposed way to study this while keeping task history the same.

We have finished the exploratory simulations. We are still reviewing the method. A confirmatory study has not been completed, and we have not established whether the idea holds over time with real people.

Available materials

What is available here

These pages contain research summaries, not full papers or reproducible experiment packages. The algorithms, experimental parameters and research code have not been released. We have not set a publication date for those materials.

A related privacy question is how an assistant could remember more about a person without sharing more of that person’s life with the network.