All you need is a VPN

it can even be a leaky one

Am only 99% sure this is true, but I figured this out in the^1 same way that Einstein discovered Relativity. Just by thinking about it.

Step 1 - use a free, cutting-edge LLM

This is cutting edge, not bleeding edge. But it is as powerful as the bleeding edge was less than 3 months ago.

With Web Search and Deep Think, the model is plenty smart and capable enough to compile information and restate it in a simplified but accurate way. Now, the competition for powerful chat + research models is in access to non-noisy data, avoidance of bias, and not getting poisoned by tarpits.

Step 2 - no IDENTIFIABLE leaks

Even self-hosted LLMs need the model to browse the web. Maybe they phone home for updates, etc. Anything you run on your machine depends on millions of lines of code and constant software updates. This makes the entire stack a highly attractive target for honeypotting of various unknown methods.

The solution is to just leak the data, bu...