MemoChat

https://gyazo.com/3ad5e109357b467a494879f9173f3c4d

2308.08239 MemoChat: Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain Conversation

https://aiboom.net/archives/54560

構造化した記憶がどのような特徴を持っていて、それがどう機能しているのか

topic, summary, range

https://gyazo.com/d8371d51aa8f07de692b6daa156188adhttps://gyazo.com/d97d7a77a685683c61a2bca627824693

できてそう

for Scrapbox

code::

You will be shown Conversation among people. Please read, memorize, and understand Task Conversation, then complete the task under the guidance of Task Introduction.

### Conversation

### Task

1 - Conclude all possible topics in the conversation with concise spans.

2- Determine the chat range of each topic. These ranges should be a set of non-intersecting, sequentially connected end-to-end intervals.

3 - Conclude a summary of each chat with brief sentences.

4 - Report topic, summary and range resutls in JSON format only with the assigned keys: 'topic', 'summary', 'startline'. Topic and summary should be in same language as the conversation.

For example, assuming an M-line conversation talks about 'banana' from line 1 to line N, then turns to talk about 'mango' from line N+1 to line M. Thus, its task result could be: {'topic': 'banana', 'summary': 'user talks banana with bot.', 'startline': '...'}, {'topic': 'mango', 'summary': 'bot brings mango for user.', 'startline': '...'}.