Kwapia: Reimagine Wikipedia by ChatGPT

Kwapia: Reimagine Wikipedia by ChatGPT

Kwapia: Where anyone can become an information creator with just a few clicks.

TLDR: Kwapia = Twitter meets Wikipedia, powered by ChatGPT.

Imagine a platform where everyone, regardless of their background, can create and share information with the objectivity, neutrality, and completeness of Wikipedia, presented with the engaging style of Twitter influencers.




Wikipedia Twitter Kwapia
Topics Narrow: knowledge only Broad Broad: Any information can be provided by ChatGPT
Presentation Long + dry Very short + engaging Short + engaging
Objectiveness Good Joke At least as good as ChatGPT
Discoverability Bad: search engine only Good: by recommendation system Good: by recommendation system + search engine
Creation <1% of users create <1% of users create Hopefully 100%
Profit Mode Non-profit + donation For-profit + ads Non-profit + ads in the form of content


Discover & read short & brief threads

Creation: Every thread is created by seeds (topics) + nutritions (prompts).

Plant (create a thread):

  1. Choose a seed (topic)
  2. Choose a nutrition (prompt)
  3. Plant your seeds with nutritions
  4. Then your thread will be available to be harvested in 30 - 60 seconds. Done.


  1. Extract: If you ever find you are interested in read or create more, simply click “Extract”: it will generate a few related seeds (topics) for you for the next round of creation.
  2. Research: Wait. What if I have my own seeds(topics) or nutritions(prompts)? You can use “Research” to create them.
  3. Transplant: But what if I don’t like all of these seeds(topics) or nutritions(prompts)? You can create a finished thread in ChatGPT and “transplant” (share) it here (Not ready yet).


Why another media? We already have traditional media, social media and Wikipedia. Let me explain:

The problem of our media:

Issues with Traditional Media You Might Agree On:

  1. Paywalls: Not everyone can get past them.
  2. Bias and Division: It’s often about political sides or what country you’re in.
  3. Young People Not Reading: Only about 5% under 24 read regularly in the U.S., and it’s even less in places like China.

Problems with Traditional Media You Might Disagree On:

  1. Too Long: 140 characters should be enough, right?
  2. Not Smart Enough: Just kidding, but who says you need an IQ over 120 to write for newspapers?
  3. Fighting Fire with Fire: They love bad/evil ways to spread news. Is using bad tactics to stop bad people/stuff really the best we can do?

Problems with Social Media You Might Nod to:

  1. Fake News Everywhere: It’s a real problem.
  2. Clickbait: Those headlines that promise a lot but deliver little.
  3. Stuck in a Bubble: Hard to see different opinions.

Community note is an answer to misinformation and it helps but it is not enough.

Wikipedia Helped But Not Enough

What’s Good:

  1. Objectivity
  2. Neutral
  3. Completeness

Where It Falls Short:

  1. Not a first choice for news: People still prefer traditional media
  2. The definition of knowledge is too narrow
  3. Too long for some cases
  4. Not personalized
  5. Passive:
    • Do you know the top articles only have a few million views a week on Wikipedia?
    • Its heavy reliance on search engines to provide answers to questions is old-fashioned.

Many enjoy reading Wikipedia, but if you ask your family, colleagues, or friends, how many articles they read daily, what would they say?

Imagine wanting to learn about the connection between Cognitive Behavior Therapy and Chinese philosopher Wang Yangming. Would Wikipedia give a straightforward answer? It can only link Cognitive Behavior Therapy more with Greek philosophy. The topic structure of human knowledge on Wikipedia is an oversimplified and rigid model and far from optimal.

Consider your children wanting to learn about Greek history; can Wikipedia offer an illustrated, engaging narrative with sound?

Have you ever considered the experiences of non-English speakers with Wikipedia? It’s 2023, so there is still a language barrier to human knowldge?

The gap between Wikipedia’s vision and human demand:

The Wikipedia aims to provide “the sum of all human knowledge,” yet ‘sum’ should imply a summary, according to founder Jimmy Wales.

It’s a ambitious and well-intentioned vision but loses touch with practicality: Wikipedia answers less than 5% of human queries.

The dilemma: When Wikipedia-level knowledge isn’t available, its impact is limited.

We often need concise information rather than long, exhaustive articles dubbed as ‘knowledge’ on Wikipedia.

A substantial market exists for social media, but it’s frustrating that most influencers lack the ability and motivation to present information neutrally, objectively, and comprehensively.

What Lies Below the Iceberg Matters, But Is Often Overlooked

Is it possible to surmount the issues posed by traditional media, social media, and Wikipedia? Kwapia is an early attempt to address these challenges collectively with AI’s help, specifically using ChatGPT in its initial version.

Tiny History:

  1. Web 1.0: In Yahoo’s era, around 1998, content was produced by professional writers for the masses.
  2. Web 2.0: From 2005 to the present, with platforms like YouTube, TikTok, Meta, and Twitter, the general populace began creating content.
  3. Web 3.0: This term has been adopted by the crypto community to mean decentralization.

The trend is unmistakable: more people are becoming content creators, and the barriers to content creation are being lowered.

Let the crypto community have their version of Web 3.0. I propose that Web 4.0 should be where every user can create for others.

Could we close the gap between content creators and consumers? Kwapia is an exploration of this possibility.

About the Name & Logo:

The logo depicts a baby capybara riding a turtle, inspired by an actual video available online.

Kwapia is a combination of ‘quark/capybara’ and ‘utopia’.

A quark is the smallest known particle, and the hope is that Twitter-like threads of a few hundred characters each could represent the smallest unit of information.

Capybaras are known for their emotional stability and friendliness, often mingling with a variety of animals, including predators. It’s hoped that Kwapia can coexist with existing media platforms, hence the preference for a .org domain over a .com, although this may change depending on financial sustainability.

Current Challenges:

The quality of content varies greatly, and not all of it meets high standards. Affording the use of advanced AI like ChatGPT at scale is challenging. Additionally, there’s the issue of content duplication.






wikipedia has an advantage where I can select the text on their website… <.<

I’m not sure whether I understand the idea correctly but it sounds fascinating. Is kwapia an AI-generated personalized facts feed? How does personalization work? I would love an explanation of the vocabulary plants, seeds, etc. What are your thoughts on personalization vs avoiding bubble formation as well on fact checking?

well, it can be fixed in one line of code. Currently it’s just a demo to better illustrate the idea, not a complete product that are consumer ready.

Is kwapia an AI-generated personalized facts feed?

Yes it is, but there are several key differences between Wikipedia & Twitter:

  1. The scope of wikipedia is narrowly defined as “knowledge”, but I hope to expand that scope to “information”.
  2. Less 1% user create on twitter and the rest simply read, but I hope to blur the line between creators and users and make more people to create for themselves and others.

How does personalization work?

It does not have any personalization yet because I just posted it here. I won’t worry about the technical side of recommendation because it is a well-defined problem and many open source solutions are available. As long as there are frequent users, achieving 80%-90% of twitter-level personalization is viable.

I would love an explanation of the vocabulary plants, seeds, etc.
I will update my post to explain that.

What are your thoughts on personalization vs avoiding bubble formation as well on fact checking?
These are two separate questions:

  1. Fact checking:

    1.1 I do think even the current ChatGPT-3.5 has been much more reliable than 99.9% Youtuber or Twitter. And this can be improved over time. But that’s just my personal impression and this topic requires research-level work to conclude that.

    1.2 I don’t think there is such thing called “absolute truth”. This is the nature of human knowledge.
    Albert Einstein: “One reason why mathematics enjoys special esteem, above all other sciences, is that its laws are absolutely certain and indisputable, while those of other sciences are to some extent debatable and in constant danger of being overthrown by newly discovered facts.”

    1.3 I think community wiki / community pruning would be an essential feature.

  2. bubble formation:
    It really depends on what topics are we talking about. If it is about “non-serious” topics like cooking, travel, sports etc., I don’t think this is an issue. Read whatever they like should be perfectly fine.

    But there are serious issues that can make a difference in people’s life and death, e.g. Palestine & Israel.

    I admit I am simply experimenting here: Would it be possible to have a PvP mode (currently it only has PvE mode meaning no direct communication with people are allowed). In PvP mode, people would generate threads for the same topic, and figure out the complete picture of complex events.

    I think this approach would be better be manipulated by biased media. But I am not sure if many people would like this approach or truth at all.

I have updated the post with explanations of terminologies like seeds, nutritions, plants etc.

1 Like

@prof.coconut Thank you!

I do think even the current ChatGPT-3.5 has been much more reliable than 99.9% Youtuber or Twitter.

From my experience, this depends on the popularity of the topic. ChatGPT has already given me hair-rising hallucinated misinformation about more niche topics such as less popular algorithms or people. But if you instruct or pre-train your model accordingly and use RAG, it might work.

Bubble formation:

Classic bubbles simply arise through typical recommendation algorithms. If I like more or less posts that problematize poverty, I might receive more or less facts about social injustice, for example.