> fully generative UIs where the HTML/Canvas are generated just-in-time
Why? The UI is a template and the core structure never changes, in the context of TikTok’s feed, so what’s the benefit you are trying to reap by dynamically generating the UI?
I thought the scroll didn’t work at first, apparently only the bottom of the screen is scrollable. I see other comments talking about a voice, I don’t have it, just the music? Also the content is really irrelevant and uninteresting (I’m assuming it uses a random article? It should aim for the most popular instead), it really needs some curation.
The subtitles are also hard to read, they should be displayed sentence by sentence, not word by word.
IMO you didn’t capture what makes some interesting content work on TikTok. The scientific content on TikTok doesn’t have the brainrot music, it’s just interesting because of the short format. It looks to me like a clone of TikTok made by someone who hates TikTok and built it out of assumptions on what TikTok is.
Honestly interesting Wikipedia articles summarized as TikTok content is a killer idea, but the implementation is not there yet
Next step is allow viewing TikTok like wikipedia. Take a bunch of popular tiktok posts, use an LLM to describe what's happening via text/screenshots (with references to memes, etc), and link them to other relevant tiktoks.
Some meme sites might be somewhat similar to this.
> Next step is allow viewing TikTok like wikipedia
I tried that and this is what I got:
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nullam tincidunt at dui vel suscipit. Donec finibus viverra tempus. Ut ut tellus ac mi ultricies fermentum a quis nulla. Vivamus urna mi, laoreet ac purus sed, ultricies tincidunt nisi. Quisque vulputate massa nec hendrerit consequat. Suspendisse potenti. Phasellus dapibus suscipit vulputate. Suspendisse id semper turpis, sit amet rhoncus nisl. Aenean auctor purus orci, eget ullamcorper lorem volutpat sit amet.
Instead of LLM and just like Wikipedia, there should be users to submit the descriptions of the videos, and see people fighting for getting the most “correct” description.
It will be hilarious to see what people will come up with when they see brain rot content.
Allow TikTok to be run like Wikipedia... So that people with no interest or knowledge of a particular subject can go round deleting said content, or blocking every shared IP from uploading content while saying anyone can edit it.
That's a very wrong view of how Wikipedia works, and frankly vastly I prefer its limited but generalized gatekeeping than "everybody is allowed to lie and bullshit and as long as it makes engagement your feed is going to be filled with it" the TikTok way.
I have at least twenty years of experience of Wikipedia, and it is exactly how it works much of the time. I lost count of the occasions when people with no knowledge of a particular subject kept trying to delete articles on it. Then there were the continuously changing rules, jargon and oligarchical structures...
Some subject areas have much better coverage than others: linguistics is much better dealt with than sport for example.
Nowadays I rarely edit Wikipedia, because they block most shared IPs. I used a named account for at least ten years and it was counterproductive.
TikTok like Wikipedia, huh? So we’d finally get to see how much of the recommendation algorithm was edited by IP addresses from US gov, Chinese gov and Russian bots?
I also thought immediately of WikiTok and was confused by this exchange and the grammar in the quote here; I thought they were claiming to have made a VSCode extension called Wikitok. I understand now!
This is amazing! I think this could actually catch on!
Plz give an easy way to follow some tags so it's not showing me football stuff. If it was just topics I was interested in, I could scroll this for hours.
This is even scalable if you consider caching some of these requests and allow users to choose if they want a just-in-time version. Also the comments really are snappy and quick to load.
747.run is a realtime infinite Ai wikipedia, every word is a link to a unique article based on the word in the URL, Ai frequently checks all URL's and writes a new article based on the word you click.
You can also write articles and all your words become links, use the website address bar to type any word after 747.run/ or type 747.run/ behind any domain to generate unique articles based on domain.
Have a group chat with unlimited people and Ai, share any link to chat with anyone, press space to chat with the Ai, have an Ai chatbot on your website or chat with users privately by embedding it on your website with iFrame (set BG color or leave it transparent)
Wow this is so freaking awesome! I guess the pages I ended up scrolling on weren't too interesting though, how do you figure out what stories to create/show?
Hi there, creator of Wikitok here!
Very happy to hear that my random little project inspired you :)
The attention to detail in the UI is good - I'm so over every default AI generated UI being rounded corners, centered divs, blue and purple gradients, etc.
Nice work.
Thanks! Half of the work here was getting gemini to generate reasonable canvas animations within the window. I'm still experimenting which "style" I should keep/add
Note there's a rate limit on generated content for the time being, why you're unable to see any new content on the rest of the feeds. LLMs are expensive!
Yea I didn't like the voice as well, it felt very coarse to me and not very enjoyable
To be really honest, I still occasionally browse shorts sometimes because of some youtubers/niche content (there is this really good etymology shorts guy and a fun fact guy I watch usually)
So like the other day I was asking fun fact about niche legal laws just out of curiosity if anyone of them might be interesting/genuinely helpful to me lol to chatgpt about different countries
There were lot of niche things but one which I found interesting was that its possible in finland and other norweigian countries to browse even private land / private forests etc. and finland even has a digital right to internet
> Finland was the first country in the world to make access to high-speed internet a legal right, establishing this policy in July 2010. The law requires all service providers to offer a minimum internet speed of one megabit per second to every household, with a goal of providing access to 100 megabits per second by 2015. PBS edri.org (from Duckduckgo AI)
A few months ago I made a (theoretically) infinitely learning geo-guessing model that updated the policy with each user guess: https://geospot.sdan.io/
Hoping to implement a simple RL loop here and optimize whats generated by the LLM to create the perfect slop machine :)
Google should buy Wikipedia and be done with it. Or we could cut out Google altogether and just search Wikipedia because it is one of the few sites Google will show you.
Google AI Studio has a Gallery[0] with some similar apps. It's an editor so you can view the code, they are usually react apps with gemini integration via genai package. Like this one here[1] is similar. It generates interesting stories to share about a route you are driving / walking / biking along. This is one of the pre-made examples I believe, I didn't make it or anything. Just to show some how some of this might work.
Yes I have base css/js that I inject on top of whatever codegen gemini 3 comes back with -- It runs via ai-sdk so the specific function is streamObject which is prompted to generate inner HTML elements
In general Just-In-Time app generation is a bad idea. The right approach is to create human-in-loop tools that a bot would recognize and invoke as needed, of course the human-in-loop tool would itself be AI generated.
The caption highlight timing is very inaccurate. It looks like it just steps through each word on a fixed timer, rather than using timing information from the TTS engine?
Yes just fixed timer, and using browser TTS nothing fancy here on purpose - when I did some research on tiktok videos generally simpler/worse quality seemed to be better XD
In this case, static text without the highlighter would have been better. Because the timing is so far off, the highlighter interferes with reading the captions for hearing assistance.
Having a generated voice tell me about some historical massacre in a chipper tone with a generated infographic and set to TikTok music was. Especially dystopian.
As an experience I found it nauseating and am never doing it again, but as an art piece I give it high marks. Good job.
wow, my attention span is terrible - the first time i tried watching a generated video, i instinctively held the right side of the screen to speed it up to 2x…
is there scope for allowing users to search for / ask about specific topics? although then you do have to think about security issues with prompt injection
Nice tech demo but in practice utterly annoying and without purpose. I mean don't you think enshitifying Wiki knowledge kind of beats the purpose of acquiring knowledge?
Deep dives with TikTok mechanics? That's not going to work, that TikTok UI was optimized for dopamine release. You want to teach junkies new things? All they learn is how to get their kicks, so outside that settings nothing will stick. When they have to apply the infotainment snippets in a real world setting the won't be these high reward stimuli. I mean if you want to hook them on feeling great while believing they've learned something...
Why? The UI is a template and the core structure never changes, in the context of TikTok’s feed, so what’s the benefit you are trying to reap by dynamically generating the UI?
The subtitles are also hard to read, they should be displayed sentence by sentence, not word by word.
IMO you didn’t capture what makes some interesting content work on TikTok. The scientific content on TikTok doesn’t have the brainrot music, it’s just interesting because of the short format. It looks to me like a clone of TikTok made by someone who hates TikTok and built it out of assumptions on what TikTok is.
Honestly interesting Wikipedia articles summarized as TikTok content is a killer idea, but the implementation is not there yet
I could see myself using that
Some meme sites might be somewhat similar to this.
I tried that and this is what I got:
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Nullam tincidunt at dui vel suscipit. Donec finibus viverra tempus. Ut ut tellus ac mi ultricies fermentum a quis nulla. Vivamus urna mi, laoreet ac purus sed, ultricies tincidunt nisi. Quisque vulputate massa nec hendrerit consequat. Suspendisse potenti. Phasellus dapibus suscipit vulputate. Suspendisse id semper turpis, sit amet rhoncus nisl. Aenean auctor purus orci, eget ullamcorper lorem volutpat sit amet.
It will be hilarious to see what people will come up with when they see brain rot content.
Some subject areas have much better coverage than others: linguistics is much better dealt with than sport for example.
Nowadays I rarely edit Wikipedia, because they block most shared IPs. I used a named account for at least ten years and it was counterproductive.
for clarification :)
Plz give an easy way to follow some tags so it's not showing me football stuff. If it was just topics I was interested in, I could scroll this for hours.
I once used it to create a "Wikipedia StumbleUpon": https://github.com/fpsvogel/wiki-stumble#demo
Here is the code related to Lift Wing: https://github.com/fpsvogel/wiki-stumble/blob/main/app/model...
Mentioning this in case you're looking for something as a basis for tags.
EDIT: Here is a list of all the categories from Lift Wing: https://github.com/fpsvogel/wiki-stumble/blob/main/app/model...
This is even scalable if you consider caching some of these requests and allow users to choose if they want a just-in-time version. Also the comments really are snappy and quick to load.
You can also write articles and all your words become links, use the website address bar to type any word after 747.run/ or type 747.run/ behind any domain to generate unique articles based on domain.
Have a group chat with unlimited people and Ai, share any link to chat with anyone, press space to chat with the Ai, have an Ai chatbot on your website or chat with users privately by embedding it on your website with iFrame (set BG color or leave it transparent)
The Ai sometimes drifts & hallucinates.
To be really honest, I still occasionally browse shorts sometimes because of some youtubers/niche content (there is this really good etymology shorts guy and a fun fact guy I watch usually)
So like the other day I was asking fun fact about niche legal laws just out of curiosity if anyone of them might be interesting/genuinely helpful to me lol to chatgpt about different countries
There were lot of niche things but one which I found interesting was that its possible in finland and other norweigian countries to browse even private land / private forests etc. and finland even has a digital right to internet
> Finland was the first country in the world to make access to high-speed internet a legal right, establishing this policy in July 2010. The law requires all service providers to offer a minimum internet speed of one megabit per second to every household, with a goal of providing access to 100 megabits per second by 2015. PBS edri.org (from Duckduckgo AI)
Hoping to implement a simple RL loop here and optimize whats generated by the LLM to create the perfect slop machine :)
https://vimeo.com/1152992073?share=copy&fl=sv&fe=ci
Day 100: Wikipedia is shut down
Do you give gemini some UI components/templates to build with or is it just prompting to get consistent results across multiple stories?
[0] https://aistudio.google.com/apps?source=showcase&showcaseTag...
[1] https://aistudio.google.com/apps/bundled/echo_paths?showPrev...
I'd like to see https://wikipedia.org/wiki/Golden_Dome_(missile_defense_syst...
https://youtu.be/h0Bg-lqNlkU
In general Just-In-Time app generation is a bad idea. The right approach is to create human-in-loop tools that a bot would recognize and invoke as needed, of course the human-in-loop tool would itself be AI generated.
Example of human-in-loop tool in use:
https://youtu.be/srG5Ze7mS7s
Seems like there's an issue with Firefox that was causing it. Chrome voice sounds "normal" - like TikTok.
As an experience I found it nauseating and am never doing it again, but as an art piece I give it high marks. Good job.
is there scope for allowing users to search for / ask about specific topics? although then you do have to think about security issues with prompt injection
Hopefully it will not make its way upstream in Wikipedia.
[Neall]