Blog post generation experiment from Scrapbox

In the context of Scrapbox automatic translation, I wrote blu3mo about the idea of "GPT could be allowed to write instead of translate" (Blog text generation instead of translation).

I figured I'd factor it out and try "Japanese" "Scrapbox -> blog post" instead of "Japanese Scrapbox -> English blog post" first.

I happened to have a bulleted list of fragments that I didn't feel like I had written everything I wanted to say, so I converted it.

An experiment in having GPT create blog posts.

code:gpt-4 prompt

Here is a memo for blog entry. Write draft in Japanese.

###

Overestimation of the value of ideas

I realized that I had a fragment of a writing after I converted it: so I came up with my own.

GPT because those feelings are the result of sticking to "what I have come up with. I saw that you generated and I thought, "That's a nice word," and wrote "stick to" back in the bulleted list.

Inside perspective, outside perspective

After converting it, I realized it was poorly worded and was interpreted differently than I assumed.

When we look at it from the inside of the self `Oh, that's how you interpret it, I was imagining the inside and outside of an organization or a community or something.

I thought supplementing this and reconverting it would change it significantly, so I split the version.

Updated ver. 2 bullet points in bulleted style based on ver. 1 blog post

ver.1 bullet points -> ver.1 blog post -> ver.2 bullet points -> ver.2 blog post

Voting is a contribution and a confidence builder

ver.1 bullet points -> ver.1 blog post -> ver.2 bullet points

A lot of text from the ver.1 blog post is used in the bullet points.

To begin with, there is an argument that an activity in the form of active reading, "itemizing a sentence that is not itemized," is beneficial.

impressions

Turning this iteration around is quite beneficial to [deepen one's thought

Bullet points are like drawings.

I just took a rough shape.

By putting it in the form of a sentence, you'll notice that you forgot to mention important context or that there are leaps and bounds.

It's a kind of "intellectual production by changing formats.

When I created this term, I was imagining the transformation from spatial arrangement to text in the KJ method.

Then active reading and Kozaneba Reading were born.

Both were human conversion processes.

With the advent of LLMs, the "LLM converts and then a human reads it" style was born.

It would be easier to get the message across if it were in the form of a sentence and then in English.

If you put it in English, it's going to be like, "You're interpreting the subject differently..." or something.

Importance of "digging" and "building up" knowledge

I wrote down the idea, then did a vector search on the results, and loaded it into more prompts.

Headings are highly constrained.

I've tried to maintain it.

---

This page is auto-translated from /nishio/Scrapboxからブログ記事生成実験 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.