
balajis
@balajis.eth
50 Following
193393 Followers
17 replies
54 recasts
300 reactions
25 replies
22 recasts
209 reactions

What does an AI-first social network look like? Some thoughts.
First, you can prototype it with Farcaster. You don’t need to reinvent users, wallets, the feed, any of that. So that alone speeds you up.
Second, open source models are now cheap enough to run that you can just put them behind a normal button. X’s Grok integration is a good example.
Third, there is tremendous room for an AI which is prompted not just on immediate context (your post) but on social context (your past posts and those of your friends). It could autogenerate prompts it thinks you’d like.
Fourth, AI improvements on seemingly simple things like image upload widgets could be profound. You could automatically search for similar images, upscale them, auto-annotate them, or do something else.
Fifth, AI could enforce a certain style in a community. As heavy-handed example, every post could be rewritten in Olde English. If you opted into this kind of moderation, it could produce unique communities. Like filters, but for text. 30 replies
35 recasts
197 reactions
56 replies
82 recasts
611 reactions
86 replies
212 recasts
828 reactions
28 replies
28 recasts
147 reactions
8 replies
43 recasts
213 reactions
3 replies
19 recasts
152 reactions
18 replies
49 recasts
256 reactions
92 replies
141 recasts
910 reactions
19 replies
40 recasts
219 reactions

$1000 Prize for AI Conference Organizer
When you’re collecting slides from many people for a conference, you need to *validate* those slides.
1) The first check is format. Is this PDF, Google Slides, Keynote, Figma, Canva? Is it a file or a URL?
2) The second check is deterministic. What is the size of each deck, the number of slides, the fonts used? Is video present, and if so is there audio?
3) The third check is probabilistic. Does the deck fit the conference format? For example, does it have a title slide? Is it all bullet points (which we don’t want) or does it have images? Each of these kinds of checks can be expressed as AI prompts.
What I want: an open source AI-based slide validator, with all the code at replit.com, which sets up a form that implements these three checks.
The workflow is: first paste in URL or upload file. Then determine format and run deterministic checks. Finally, run each AI check as an individual prompt. The result is a list of ❌and ✅ for every unit test. @bountybot 13 replies
93 recasts
299 reactions
Cloud Cartography: $1000 Prize
We need maps of the cloud. And Farcaster can start providing those maps, because it's an open state social network. That is: Farcaster's database is public and viewable if running a hub.
So here's what I'd like to see: a Python script, hosted at replit.com, that lets you paste in N usernames to generate an in-browser visualization of the social network between those N people, with a slider that visualizes their interconnections as a function of time.
As you drag the slider, your code should display various metrics like the number of edges, the adjacency matrix, and the all-pairs-shortest-path matrix (https://coursys.sfu.ca/2020fa-cmpt-307-d1/pages/slides21/view) as a function of time.
This will allow you to visualize a network of N people coming together, by connecting, over time. All your code should be open source, your visualizations should be pretty and mobile friendly, and you can add other features so long as you nail the main ones. @bountybot 23 replies
144 recasts
443 reactions
35 replies
83 recasts
516 reactions
9 replies
33 recasts
141 reactions
14 replies
17 recasts
125 reactions
14 replies
6 recasts
72 reactions
27 replies
32 recasts
170 reactions
18 replies
14 recasts
94 reactions
20 replies
7 recasts
96 reactions