Xtr3m3hodl

Just implementatwd a function initially in JavaScript that took 148048ms in JavaScript and the rust variant ported to warm took 3.93ms.
Apps and servers are about to get a lot faster because systems languages are now approachable thanks to LLMs

2026-03-24 07:50:40 UTC

Wisp is pushing boundaries when it comes to nostr clients. Just observed how replies are guaranteed as much as possible to be broadcasted to more relays. @nprofile…jejf this is really impressive.

2026-03-24 07:45:33 UTC

- reply

Yeah really great animations and storyline

2026-03-24 05:27:40 UTC

Computers are really fast but because the vast majority of applications today are built with dynamic programming languages and the rush to moving fast and breaking things championed by VC, have blurred the line so bad that now we have slow programs
With systems language like rust, c, zig etc, simple binary search operations for example over a large space can run so fast that attempting to do this in dynamic languages would just not be practical.
The vast majority of all the fancy algorithms become almost unnecessary because these compilers have been optimized for speed.
This is probably the best use case for LLM imo, because these languages are hard to learn and not the best for rapid prototyping or thinking about new features. But the moment a feature or prototype or MVP is in place, for the sake of stability, switching the high impact areas to these system languages is always worth it

2026-03-22 10:46:38 UTC

- reply

Especially when the delivery is to be consumed/used by other people and not one self. Validation is still a nightmare

2026-03-21 08:49:22 UTC

Codex is way better than Claude code at following instructions and maintaining existing codebase conventions. Can't believe how codex is just like reading my mind when it comes to fixing issues

2026-03-13 15:15:54 UTC

- reply

We need to change your title to nostr benevolent commander. webmaster doesn't do enough justice.

2026-03-04 05:29:38 UTC

- reply

the app is really good and snappy. it is my daily driver now.

2026-03-04 05:06:23 UTC

- reply

What I end up doing in these scenarios is to ensure that I have ssh-copy-id setup to the vps then get Claude code to ssh and fix the issue. Did this recently with a proxmox container running openclaw

2026-03-04 04:57:16 UTC

- reply

It is like a vps but it's on- demand (no ssh setup necessary since all that is abstracted away in d cli tool). You have the ability to spin up and shutdown/pause the sandbox. Its just another option among the ever growing list of running agents in a sandbox.

2026-03-03 22:41:00 UTC

- reply

I completely abandoned it as well. Ended up using either sprites.dev or cursor agents with my cursor subscription

2026-03-02 18:15:33 UTC

It is appearing the war is probably an excuse/distraction over the Epstein files

2026-02-27 10:59:46 UTC

- reply

Is the one person solving just their problem or problem for others because there lies the difference.

2026-02-25 19:11:00 UTC

- reply

One would think this would be obvious. Chat interface isn't always ideal. UI is still really important for the vast majority of users.

2026-02-18 14:15:51 UTC

- reply

Until they straight up have amnesia in between. Had an experience yesterday were Claude forgot about the task it was doing and had to remind it.

2026-02-15 00:10:25 UTC

- reply

Are these personal apps for single individuals or apps used by other people because this is just not the case.

2026-02-13 14:50:39 UTC

- reply

You are not. Ai content in small doses is fine. When it is all I see, it's time to mute all traces of it

2026-02-13 14:35:25 UTC

- reply

This was just brutal

2026-02-11 19:21:31 UTC

https://youtu.be/mb5Lx4auBKI

2026-02-08 04:41:23 UTC

- reply

Same. Got it to do a refactor 2 days ago within a specific scope and then yesterday it just botched similar patterns completely.

2026-02-07 23:22:40 UTC

- reply

Kryptonite was so cool

2026-01-27 06:55:06 UTC

- reply

Ai is never learning how to write memory safe c with the way it keeps forgetting instructions on occasions. Rust seems to be the best option for AI since it can keep fighting with the compiler.

2026-01-15 01:48:05 UTC

- reply

Gpt codex 2 for coding through cursor.
Opus 4 for planning and general troubleshooting as well as iterative development

2026-01-04 23:29:11 UTC

- reply

Tried opencode? Imo, a better more polished alternative with the ease to switch between different providers.

2025-12-10 12:10:23 UTC

- reply

💯. No stress involved and the trust element does help a lot

2025-12-05 11:03:09 UTC

- reply

The third option is just walk away 😂

2025-12-04 12:24:57 UTC

- reply

One battle after another.

2025-10-30 17:45:52 UTC

- reply

South park has a recent episode on this trend. Pretty hilarious 😂

2025-10-21 21:20:35 UTC

- reply

Probably because it is a different twist on the medical procedural drama. It is kinda like 24 but for emergency medicine.

2025-09-29 14:46:22 UTC

- reply

You have a gang waiting to join with this class action

2025-09-26 18:43:51 UTC

- reply

Long term reputation loss for a brief moment of viriality

2025-09-17 18:41:19 UTC

- reply

I didn't appreciate this anime until years later.

2025-09-17 15:00:37 UTC

- reply

It gets better in every episode. Would not provide any spoiler going forward 😂

2025-09-17 14:45:51 UTC

- reply

The latest Manga episode from last week

2025-09-17 13:30:58 UTC

- reply

Shot a celestial dragon when he was still in the Marines

2025-09-16 19:14:03 UTC

- reply

Just like his dad from the last episode

2025-09-09 14:29:04 UTC

- reply

This so funny on different levels 🤣

2025-09-08 20:55:07 UTC

- reply

We are probably arriving at the point where third party dependencies would need to be hosted locally with running code to limit messes like this. Supply chain attacks are the worst

2025-09-08 20:35:20 UTC

- reply

It is hard to keep a level head and think critically when the algorithm incentives mobbing. One consolation I take from scenarios like this is appreciating there are still critical thinkers out there and follow them.

2025-09-05 12:03:54 UTC

- reply

I don't get how lucrative spam is to be honest in this context.

2025-09-04 15:49:11 UTC

- reply

This meme should should be framed everywhere

2025-08-31 16:38:02 UTC

- reply

Keychat greatly helps with this and does support amber so there really isn't any excuse not to

2025-08-29 23:19:25 UTC

- reply

The beef we didn't ask for but needed 😂. Robot slaves vs child labour in sweat shops.

2025-08-29 03:25:27 UTC

- reply

The vlogs are working. Gradually then suddenly.

2025-08-27 00:14:39 UTC

- reply

These series are fun to watch. The intro 😚

2025-08-26 23:49:56 UTC

- reply

Okay this is really fun. Nostr is inevitable eventually

2025-08-26 12:59:12 UTC

Nostr as an options market
#nevent1q…fp49

2025-08-26 12:58:28 UTC

- reply

$115400

2025-08-19 11:33:55 UTC

- reply

Could this just be skill issue 😜.

2025-08-02 16:39:26 UTC

- reply

After setting up an identity, use the in app browser to visit a nostr app like yakbak.app for example and login. https://blossom.primal.net/ae196483db3ae0dd7a1a047e8a21317ce2280e9602792ccf70bccaaaef975b42.jpg https://blossom.primal.net/1ad3f6754a9d394b019071081839d80b476b1ca1e1691d219cb4200d6b3c9600.jpg
You can select any of the identities you have created and you should be logged in. Your nsec is local to the keychat app just like a browser extension

2025-08-02 16:35:30 UTC

- reply

It's all good. You can create a new identity by linking your existing nostr nsec https://blossom.primal.net/20fca027c480a95cecab2f0b63f542260c01c9769faa718a0552b8e6a8834a72.jpg
You only need to do this once and then it has an inapp browser to test other nostr apps using the same browser extension setup that works on desktop devices

2025-08-02 16:20:30 UTC

- reply

Try keychat. A safe way to test different nostr apps without compromising your nsec

2025-08-02 13:52:19 UTC

- reply

Same as well I always wished I gamed more but just can't justify the time. Yet I can justify the time spent building apps 😅

2025-07-31 17:14:34 UTC

- reply

Yes you must 😂

2025-07-21 00:01:51 UTC

- reply

For lex Friedman 3x . I can't stand the dragging

2025-07-20 18:35:15 UTC

- reply

It was beautiful to listen to at 2.5x speed because of the length 😅

2025-07-20 18:21:02 UTC

- reply

It is a private project but I could send a GitHub invite .

2025-07-20 18:15:36 UTC

- reply

In my experience, it is the opposite. After I have setup some structure and convention, the LLM just goes in full beast mode.

2025-07-13 08:17:26 UTC

- reply

Claude code's plan mode would get you pretty close.

2025-07-08 18:32:25 UTC

- reply

So basically vibe coding for native development instead of web development?

2025-07-05 12:16:38 UTC

- reply

But actions caused it to arrive at this point. It didn't get to this out of thin air but as a result of actions of individuals. So at the end of the day, the integrity of the person matters irrespective of the timeline

2025-07-01 00:46:59 UTC

- reply

Don't you think it depends on the LLM provider? And when you say better, it isn't obvious what you mean since one would imagine the types on typescript would act as a feedback loop to the LLM when something is off?

2025-06-30 19:50:18 UTC

- reply

😂👌 beautiful

2025-06-28 04:02:47 UTC

- reply

Just saw this in my mail as well. Checked the sender id and found out it wasn't from trezor.

2025-06-27 10:04:43 UTC

- reply

Even when working with others, and especially on existing projects, that's how I mostly work. I'm still familiar with all the obvious software development practices. It is just, I don't dwell too much on a written well thought out plan (things change more often). The way I see it, the "plan" is being developed mentally as I attempt to solve the problem

2025-06-25 15:33:26 UTC

- reply

This would be a great feature if eventually supported

2025-06-21 13:54:30 UTC

- reply

The best part of this talk IMO was about building more augmented systems and less autonomous systems

2025-06-17 18:45:12 UTC

- reply

Oscar worthy story telling 😂

2025-06-16 17:37:42 UTC

- reply

I guess I'm not too shocked by this. Any company in the ads business is probably doing something similar. It's a good thing I don't have Instagram or Facebook installed on my device and haven't visited the site since 2017.
The native app is basically running a background service that the web pixel connects to. So it isn't really a problem with web browsers.

2025-06-16 16:39:45 UTC

- reply

This requires having the Facebook mobile app installed though which is responsible for running the background service. If one doesn't have the Facebook mobile app installed, this doesn't work the way it is described?

2025-06-14 02:50:06 UTC

- reply

Some of them want to use you

2025-06-13 17:40:27 UTC

- reply

This is really cool.

2025-06-12 15:42:06 UTC

- reply

I think he meant the kinds nip convention 😂

2025-06-11 03:19:35 UTC

- reply

Was thinking about this and the idea I came up with was some sort of API that connects to something like audit.8333.space (extended) to validate a specific mint. Just like with mempool.space

2025-06-10 16:22:09 UTC

- reply

Your uncanny story telling ability is just 🤩

2025-06-10 12:19:12 UTC

- reply

Unless the ai model runs natively on your device. If not you are always making API calls to the LLM provider which renders the ability to run offline apps impossible.

2025-06-09 22:57:06 UTC

- reply

Isn't the price the same as 3.7?

2025-06-08 03:27:16 UTC

- reply

Try out #keychat. One app to test all the mini web apps with support for multiple accounts and different login methods.

2025-06-08 03:25:09 UTC

- reply

They support amber as well. https://blossom.primal.net/bd68a4e39778bdcebf458ce595a442b887971982948d96a732d192c99c27d500.jpg

2025-06-08 03:20:47 UTC

- reply

You could test it out with a test nsec to actually see how it works. We all inserted our nsecs in amethyst before amber became the norm. They are actually moving in the right direction

2025-06-08 03:04:00 UTC

- reply

I meant as a browser for mini web apps? They support nip07 which makes it extremely easy to test out a lot of web apps without the inconvenience of bunkers. They have gone a long way from just a messaging app

2025-06-07 14:50:08 UTC

- reply

And it is functional. Can only send funds if the device is on the tailnet.
https://blossom.primal.net/3a34abdcfce24be199271f1cb08e8a6c566f7969f6e0859eb2bde8e32165cdb8.jpg

2025-06-07 14:21:32 UTC

- reply

About to test this with albygo to see if I can access the funds wallet over tailnet

2025-06-07 14:20:48 UTC

- reply

The idea is to add an additional layer of security to the generated connection string. The user would only be able to connect with the connection string if the device is part of the tailnet.
If the device is stolen, just removing it from the tailnet prevents loss of funds

2025-06-07 12:49:42 UTC

- reply

I am looking into only exposing the wallet relay on a tailnet

2025-06-07 10:41:59 UTC

@nprofile…6u4e does haven support nip-47?

2025-06-06 03:47:58 UTC

- reply

This realization kicked in today when I used Claude opus which is supposedly their best model to come up with some sort of structure. It did a good enough job but because of my paranoia, I began going through each files created and began noticing weird cases that were either inefficient, unnecessary or just dangerous. Even when asking it to generate tests for the work it did, couldn't differentiate when mocking was necessary and when it wasn't. It isn't as rosy as it is being painted.
In the bright 🌞 side, I did learn about git trees and would be using it more in my projects.

2025-06-06 03:03:55 UTC

- reply

I'm not so sure about this. LLms mostly learn from data. If the vast majority of data that would be available from for example the current point LLM generated, then there is no new insight to be gained.
I'm not against it's use but can see its limitation and ceiling that can't be broken because of how it's fundamentally works. If another branch of AI that isn't LLM based comes out, that's a different story but with LLMs, there is a ceiling that just can't be surpassed. And that might be good enough for most. Just something to keep in mind.

2025-06-06 02:47:39 UTC

- reply

I think we are seriously overestimating the power of LLMs especially when it comes to building apps. Even the best models introduce subtle errors that can go unnoticed if you aren't paying deep attention. Just asking the model to generate tests should be scanned extensively with a fine-tooth comb.
if the goal is to build a throw away prototype sure full steam ahead. But for something that would be maintained and has more reach outside just self usage, this is a disaster in the making if not extra vigilant.

2025-06-05 14:44:52 UTC

- reply

Loving this so far. Had some issues initially using opus but seems to be resolved. Nothing beats paying with lightning

2025-06-04 17:12:00 UTC

- reply

I think there is a YouTube channel that does something like this https://youtu.be/ZB4soi4FZJc

2025-06-04 16:24:38 UTC

- reply

This song is banging 😍

2025-06-03 15:32:44 UTC

- reply

This is really great. The only problem I have is if I have to be this detailed all the time, I might as well just right the damn code.

2025-06-02 17:53:39 UTC

- reply

💯 came to the exact same realization early this year. Just need to rip the bandage

2025-05-31 15:53:32 UTC

- reply

I think this is probably the easiest but just minor coding https://bitcoin-connect.com/

2025-05-31 09:51:28 UTC

- reply

From my little sleuthing, one of the accounts following the bot is followed by btc sessions which I think follows Derek

2025-05-31 09:48:04 UTC

- reply

This 3 are on the utxo's web of trust.