Could Saline validators be replaced by ChatGPT?

Mar 19

A thought experiment

The core idea behind Saline

In Saline, logic is installed on assets. Transactions can then be presented to the assets, and validator nodes determine whether each transaction passes the logic or “mandate” that gates the asset. If so, the validators vote to include the transaction.

In practice, this involves writing an interpreter which is then run by each node to evaluate the logic. Could we replace this interpreter with an LLM? Let’s try.

Architecture

Each node will now be a ChatGPT (or Llama, or whatever LLM you prefer) instance which will accept the transaction and its mandate, and determine whether the transaction should be accepted.

The nodes then vote to accept or reject the transaction (for simplicity’s sake, let’s go with BFT’s 3n+1: 7 positive votes out of 9 nodes are needed to proceed with the transaction).

A simple example: limit order mandate

ChatGPT seems to be a fine interpreter for a simple price-based condition! Perhaps Cowswap should look at it for the evolution of the CoW Protocol.

A mandate with policy functionality: whitelisting and conditions

ChatGPT did very well in evaluating a policy mandate. Watch your backs, Fireblocks and Predicate.io!

What about dynamic vesting?

We previously (https://www.saline.network/blog/dynamic-token-vesting-with-intents) discussed the idea of vesting investors as a function of how the chain is doing. Let’s see if ChatGPT can implement the vesting part of the mandate (for simplicity’s sake, we are not yet including the market making part):