Thoughts on what could we get if we distilled the educational discovery aspect of Twitter into a separate system.

Problems⌗

There’s a lot of valuable yet hidden information on the internet⌗

Google’s response quality strongly correlates with the quality of the question. Oftentimes we don’t have the right question: either because we lack understanding on the topic to formulate it, or because we didn’t think that we needed particular knowledge until we learned it.

We procrastinate anyway so why not make it useful⌗

I believe that such an app could become a good replacement therapy for people who get used to open their social network app when they feel anxiety. I used to do this myself with Twitter and after switching to Youtube which is trained on STEM topics I’m interested in a year ago I feel much better.

Solution⌗

Resource summary as a content⌗

It’s always a wonder when you accidentally stumble upon a twit where someone mentions an article on a topic relevant to you right now. Imagine if your whole feed would consist of such twits: summaries of relevant resources with a link to the original. Resource could be any form of educational content: an online course, a blog post, a research paper, a YouTube video, a podcast.

Personalization⌗

There’s too much information on the internet, so we need a way to filter the relevant one. There a couple of ways to do that:

  • suggest user to make all relevant search requests through the app and
    1. either redirect them to Google, logging the request
    2. or search through the database of already generated summaries
  • allow to like or dislike twits
  • allow to follow topics

Interaction with AI⌗

Users will be able to ask AI questions, request feedback on their work or maybe even discuss stuff just by tweeting, as if it was a regular Twitter.

[bonus] Many competing AIs⌗

As opposed to the Twitter fighting bots, we could actually encourage people to make AIs which compete between each other for being useful to the user. As an incentive system we could split user subscription fee in proportion to the amount of positive interactions with the tweets generated by an AI.

[bonus] Every person has a unique tuned model aka avatar⌗

We can think of the model trained on our preferences as of an avatar of ourselves, which is able to constructively communicate with such models of other people and come up with new ideas.

[bonus] What would Einstein suggest on this topic?⌗

Fine-tuning models on the data we have of prominent scientists, we could essentially talk to the dead and potentially get valuable insights. Though we would need a lot of accurate data, which is especially hard to come by for the people who lived before the invention of the Internet.

[offtop] Simulating events with models tuned on communities⌗

Fine-tuning models on replies of specific communities can be handy in learning their opinion or predicting the impact of potential actions. We could train the model on the “smartest” people on Twitter and ask what they would say about a certain article or event. Or we could simulate the reaction of specific groups on geopolitical events.

Users⌗

IT Twitter/Tiktok crowd⌗

It’s not a replacement for social platforms, but some people might find it better for their mental health: no drama, no tribalism, no hate.

Youtube learners⌗

Youtube has a great recommendation system for learning. The only drawback there is too little content. There still appear to be much more content in the text form as of 2022.