A novel UI for exploring nested topics using LLMs to generate deeply nested "concept" trees

Expand the tree conditioned on the previous branches

Edit branches to guide generation

The main novelty here is that generating tree branches is done through a single prompt and using its ancestor branches as context

SUBTREE_V2 = """
Given the following background context, complete the task:
BACKGROUND CONTEXT:
{context}

TASK:
Your task is expand on the following subtree. Here are some requirements:
- generate 3 new levels (child nodes) for each node in the SUBTREE nodes
- use the ANCESTOR nodes as context but do not generate any new levels for it 
- never change the root of the SUBTOPIC nodes
- only give your output in terms of the generated subtree
- ALWAYS start your output with "[0]"
For example, given below:
ANCESTORS
--------
[0] Russia's situation leading up to the Great War
[1] Social unrest
SUBTREE
--------
[0] Role of the intellectual elite
[1] Advocacy for political reform

The new output would be:
[0] Role of the intellectual elite
[1] Advocacy for political reform
[2] Creation of liberal publications and newspapers
[3] Disseminating propaganda among workers and peasants
[3] Promoting political awareness and activism

Here is the subtree:
{subtree}
======================
"""