Coding 🤜🤛 Nostr
Building a private AI assistant for truly private chat and personal data processing (e.g., notes, PDFs).
Planning to create a PWA web app that allows users to run any LLM model, including the latest LLaMA 3.2-1B, locally on any device or OS, even offline.
Users will simply select the model, and the app will automatically download, cache, and store it for future use. The next time, loading will be much faster.
The MLC-LLM library makes implementation straightforward, supporting all platforms and leveraging WebGPU for maximum performance.
Coming soon, meanwhile check there library repo. They have lots of examples.
MLC library repo: https://llm.mlc.ai/docs/deploy/webllm.html
Got "file upload" working with the latest Llama models.
Also, it seems the smallest Llama 1b (fine-tuned) model performs even better with more context.
Which is really good because that one is really fast even on a medium-spec device. 
Opinion: French isn't a real language. 
Building a private AI assistant for truly private chat and personal data processing (e.g., notes, PDFs).
Planning to create a PWA web app that allows users to run any LLM model, including the latest LLaMA 3.2-1B, locally on any device or OS, even offline.
Users will simply select the model, and the app will automatically download, cache, and store it for future use. The next time, loading will be much faster.
The MLC-LLM library makes implementation straightforward, supporting all platforms and leveraging WebGPU for maximum performance.
Coming soon, meanwhile check there library repo. They have lots of examples.
MLC library repo: https://llm.mlc.ai/docs/deploy/webllm.html
I dedicated nearly half of my weekend to this project, and it's undoubtedly one of the most ambitious ones I've taken on so far.
Building regular lightning paywall services is relatively simple—I built one 1.5 years ago that's used by thousands today.
However, building one without any middleman (no centralized server, no LNbits, no personal node, not even a central database) is an extremely difficult challenge.
Here's what to expect:
Seller side: Your content, price, and wallet address are all stored in a shareable QR code, secured by cryptographic encryption and signatures.
Buyer side: When the QR code is scanned, it initiates the payment, and after it's completed, the content is decrypted and displayed.
For Developers: Ideally, you shouldn't even need an API. You should be able to implement this functionality by simply importing a library with a one-liner, like:
generatePaywallQR(contentURL, price, lnAddress)
I'll be working on it this week, and hopefully by next weekend, I'll open it for some beta testing. After that, we can keep improving it based on feedback.
This doesn't even account for X's new 'speech but NO REACH' policy, where "controversial" accounts are completely nuke from public timelines and replies, even within their followers' feeds, which are now completely algorithm-driven (TikTok-style).
This can't be the future of online communication. Nostr is the only solution that addresses every issue mentioned.
If you're paying for X to support free speech, I have a bridge in Brooklyn to sell you. 
Podcasting bro 😭 
nostr:nprofile1qqsvdac80utfn4gvly4fv54la0l6cp0udpptnm3ezzyajkdc44w53lgpzemhxue69uhhyetvv9ujuvrcvd5xzapwvdhk6qgdwaehxw309aukzcn49ekk2qghwaehxw309aex2mrp0yh8x6tpd4ehgu3wvdhk68na3xp and unless I'm missing it, perhaps a delete function on long press or check selection for urls or notes?
Otherwise slick for application selection. 🚀
Swipe for detete, in both notes and URL.
App is bit rough right now. I haven't released it yet, but the next version should be better.
Test new Llama 3.2 1b/3b on any device locally with the web app.
I have multiple apps that use this WebGPU implementation. Decent small LLM models were the only constant, but now we finally have the model.
One of the Greatest!
Llama 3.2 3b fine-tuned model running locally on device offline, at around 10 tokens/sec. 👀 
1B & 3B models 👀
These should be able to run on iPhones models going back 5 years or so to around iPhone X 
I used to run a node to power my LNbits backend for my Lightning paywall services (Zapit), but the new version no longer requires a personal node or LNbits, so I don’t run it anymore.
Personally, I hated working with LNbits and running a node, especially for build commercial stuff & services.
First step is complete. Now, we are extending this isolation to all the apps pinned in each AppCore variant.
You should be able to log into 3 different Signal accounts using 3 AppCore variants.
Each variant should behave like an isolated environment. 
Nostur experience on iPad 🤍
It's my fastest and best multitasking experience on iPad so far. Perfect for both power users and casual users, it's very easy to navigate with all the necessary features.
3 key features in upcoming AppCore V2:
1. Create multiple copies of the AppCore app & run them simultaneously, separate from each other.
2. Categorization of apps with custom Profiles like Personal, Work.
3. Database will now be encrypted with password, in addition to being completely offline and local.
If you're looking for an app that can help you organize all your Nostr Native apps and web applications/websites in one place, complete with an app launcher and a native custom tab browser, then look no further! 🤜🤛
New service that converts complex Arxiv research papers into easy-to-follow audiobooks! 🤌
>Redditor posts picture of his brain MRI
>Large portion of his brain is missing
>Several other Redditors chime in with their own MRIs
>They're all missing large portions of their brains
It's all starting to make sense now. 
Everyone has become so comfortable with scams, from politicians to creators. They do it in broad daylight, shamelessly taking advantage of the most vulnerable and desperate people. It's very disgusting behavior, and it should not be normalized.
BlueSky is also comes in almost every discussion about X alternatives with thread. Thread is partially opening up to the fediverse. Interestingly. Mastodon has some very great and legendary users, such as Terence Tao. Nostr has a long way to go.
coldest moment in mcu/dc 🥶
https://video.nostr.build/9df6296a15df81ed938f77b77943bc068b010cdfe6df53ae8013ab429ecd364f.mp4
It might take some time, I'll start by rolling out the paywall feature, and then I'll speedily work on adding support for physical items and subscriptions.
Yea pretty much. There's an example service out there with LN and NWC for subscriptions, called Flash, though unfortunately it has a lot of steps to get a payment going, and they're the middle-man with a fee you're trying to get rid of, but it should serve to help as a reference for what you're building. Here's an example subscription page I made for testing purposes a few days ago: https://app.paywithflash.com/subscription-page?flashId=676
(Make your own and see how they're doing it. I'd say there's definitely a lot to improve with their UX / user-journey flow).
Aside from that, you could also look at how traditional sites do things (a popular one being Patreon, I like SubscribeStar from what I remember, and a bunch of people use ko-fi).
(side note: if you'd like, since it would potentially end up with me benefiting from it as well, I'd be more than happy to help with design/ux/user-flows, if you're wanting and willing to accept the help/contribution and talk it over, this or other aspects of it as well).
Once I have the core functionalities implemented, I will send you a development version URL so you can monitor improvements in real time. I will also be open-sourcing the code, making it one-click self-hostable, And easy to modify.
My answer might be tricky. I don't want the current AI architecture (Transformers) to define the future of LLMs or ML, that would be disastrous. We need a breakthrough!
It's not that I can't live without it, but AI has countless potential use cases across fields like computer science, medicine, and space exploration. It's already penetrating every industry, these new generative models being just a part of it.
That's why I believe it's crucial to shape AI for a better, less centralized future.
What amazes me is how the open-source community has kept pace with frontier models. If OpenAI disappeared tomorrow, we could still rebuild up to GPT-4o using just open-source models.
I hope it makes sense 😅, I'd love to hear your thoughts.
Thank you for taking the time to provide such comprehensive and useful answers.
If I understand correctly, it should be something like this preview page, but with some modifications, such as multiple picture uploads for showcasing, markdown support for descriptions, and multiple subscription options, simple login. 🤔
It should include proper and elaborate feedback, perhaps through Nostr DM, optional email.
And dark mode, of course
I'm rebuilding everything from scratch, aiming for a minimalist approach to selling physical products.
Something like, it will include product previews with long-text descriptions and markdown support, as well as multiple image uploads for showcasing. There will be extra input fields for user addresses and other necessary details, which will be forwarded directly to your Nostr DM and possibly via email.
It's entirely feasible; I just need time to get it right. This time, I want to perfect the monetization service.
You will be the first early tester, I will really need your suggestion and recommendation.
I hope more people start using it. I use it all the time with my AppCore app using the '!chat' tag.
https://video.nostr.build/1ce1f6983e8a7e48ad280b7998ae9bb41430c04d13722aef4188e3d8e1744722.mp4
We must liberate AI ✊ 
Zapit Development Summary | 80% Complete
I've added a direct LN wallet proxy, eliminating the need for Lnbits.
Additionally, I've implemented the LNURL LUD-10 protocol to package encrypted content, LN address, and a price into a single invoice, eliminating need for a centralized database.
Thanks to Alby's excellent work, I've integrated NWC for recurring payments & subscription based service.
If everything is implemented correctly, you won't even need my API. You can integrate Zapit features into your app using a simple library, no permissions required.
This version of Zapit is a complete rebuild, designed from the ground up with best practices in mind. It's one of my most ambitious projects, and I'm not looking to gain anything from it.
My goal is simply to create something I can be proud of.
I understand that many Nostr users dislike paywalls. I respect that viewpoint, but I also believe that artists in the Bitcoin space deserve to have these option to monetize their work.
I'll release an early version of Zapit soon and will continue to improve it based on user feedback.
My previous app,' Zapit' , used the same approach. However, new version eliminates the need for Lnbits.
The problem with Lnbits is that if you're not running a self-hosted instance, the Lnbits server owner can delete, modify, or redirect your transaction. If server is down you can't withdraw.
In previous Zapit, I was the one managing that server, it's a mess but this new version takes a different approach, this one is directly connected to your wallet.
Isn't it wild? A lot of that money isn't even theirs, but they still use it to control and buy practically everything and everyone.
Even with complete uncensorship & optional anonymity, Nostr still has the most wholesome content. How?
Yeah sometimes Perplexity, most of the time I use Arc search for complex web queries; I like it's UI, It's handy for getting a quick overview or doing some light research.
However, I haven't really done any serious work or extensive research using these models with internet access.
https://video.nostr.build/184dbbe8bf2ac63e334cba639ded82290da46daf16d6a2530c52700a99517e00.mp4
Now they are making math illegal, you can't perform mathematical calculations on your phone, you need State's permission, wow 