Open briansunter opened 2 years ago
Thank you for the answer. But the Contest of the text can change if you cut the text in the wrong Position. You know what I mean?
It is true that it loses some context, but I've been surprised at well it can still get the meaning even with the chunks.
One other idea is feeding the summary of the first block along with the full text of a second block.
Unfortunately GPT-3 is not the best for text summarization, since they limit how much text you can send them and it's pricy. But if you break it up into chunks and do some workarounds you can still get decent results. I'm looking into #13 for this too.
thank you!
What about Readwise Reader's approach? If too long basically break it up into central sentences.
Ex:
Give me a list of the key points of the article "{{ document.title }}":
"""
{% if (document.content | count_tokens) > 1000 %}
{{ document.content | central_sentences | join('\n\n') }}
{% else %}
{{ document.content }}
{% endif %}
"""
Limit the length of text it can send.
Short term give a warning if they send too much. Long term, send it in batches.
Maybe use #13 for long text summaries.