divyenduz

Thanks for the comment, author here "losing the community aspect of it", this assumes that I only use LLM for that, and that's not true. I am involved in communities via both Discord servers (recommend the one from The Zone simulator) and local meetups too.

"lot of knowledge on the websites, freely available, for anyone who wants to read", of course, I read that a lot as I mentioned in parts of the blog post, I didn't want to out source "learning" and I did many things the hard way i.e. reading + understanding and continue to do so.

LLMs were mostly useful in two things: - I could make a copy of my apartment (it is much more accurate by now) and I can practice simulator and real life in the same place. This is why the creator of The Zone built it too. This without AI would have been a multi week project for me and with AI it was a 2-3 day on the side thing. - When debugging settings things up, I can give it access to read the real data from my devices, videos. No online reading has that context and can translate that into insights quick. It is more about speed than AI.

I also mentioned "what's learning anyways" at the bottom of the post, FPV flying is a good example because unlike say writing content or software or creating something the AI can't do it (it can fly drones with IMU, GPS but not tiny whoops) for you and can only assist you.

Would be curious about your blog post on the subject btw. Won't mind you sharing that.

A tinkering experiment written with AI

https://x.com/divyenduz/status/2010959045436260795

Testing the boundaries of what I can do with AI and how to prevent AI slop and produce meaningful things 100% hands off.

Is this meaningful? I don't know yet.

Is there a youtube channel or something that does deep dives into antique source code like this or windows XP?

Plug, wrote something along these lines. It is a FUSE file system and the storage is SQLite

https://github.com/divyenduz/zoid-fs

Oh nice, also check out https://github.com/divyenduz/trackfootball.app

It is like Strava for Football (gets data using Strava OAuth). Open source, but docs aren't simply there yet for host-ability!

Also, don't want to hijack your post, but some people might be interested in a football variant! Will defn try Endurain soon

Oh nice, I did something similar recently (https://github.com/divyenduz/zoid-zip), mine is much lighter/smaller than yours. Instead of following a standard fully, I just use Huffman coding to make stuff smaller!

Sounds like a good use case for TailScale funnel. Not sure if this is possible today but can work via that route.

I have done something similar with Raspberry Pi and Tailscale. Really happy with the setup. Almost 6 months in and works like a charm.

https://twitter.com/divyenduz/status/1597863894055518208

Sure, added the docs here https://github.com/divyenduz/backup-scheduler#restore

Basically, you can just manually call the sync command with s3 bucket and local directory arguments in reverse or use aws s3 cp --recursive

This is very nice, I wrote something very similar for telegram, I don't maintain it anymore, love to see tools like this

- https://github.com/divyenduz/languagelearners - https://languagelearners.vercel.app/

It will be hard to get this working on with professional teams as is, you see wearing a watch is not allowed in professional sports.

Besides, this is almost a solved problem in B2B, many many provides for professional teams

Hey folks, created this project to track my Football/running activity.

No real football games to track yet because of the pandemic but I did track some running/kicking the ball sessions.

Here is how they look like - https://mylifefootball.com/activity/46 - https://mylifefootball.com/activity/4

Very very early days and it is something ready only for power users. Login with strava is required and the "social" aspects of this might even violate their API terms (so, If I pursue this further, I need to create my own apps collecting data)

How does this look like? Anyone else besides me interested in something like this.

I made this a while ago, the primary requirement was to build a static blog but without writing code/components and hence I used a WYSIWYG editor.

I would at some point use TinaCMS in place of this custom editor I slapped together.

Doing something very similar, hosting a lot of things on a Raspberry Pi 4 with 400 GB SD card.

Dockerizing most things https://github.com/divyenduz/dev-infrastructure

Not as easy though, I still need to figure backup strategy and everything. My goal is to eventually remove photos, and almost everything hosted entirely really.

I agree, would be an interesting POC, on my backlog

Performance? Not sure TBH, wasn't a primary aim for me. The slowest bit of this experiment is to load the wasm file in the browser.

But that’s a bad idea, once browser version will be locked to a database version, security updates etc would be tied and a yada yada

Ha, would be nice-ish if browser have this pre-available, TBH most loading time is loading SQLite async, then it is instantaneous

Ha, what’s your aim with that? Making changes in the browser and have them saved somewhere?

I am pretty sure I can make this upload to s3 or something to emulate that

But a part of this experiment was its offline nature, where you can play around with tweets etc without having them uploaded to a server.

Interesting idea though about having them “sync”

Aha, amazing, I have played with datasette in past, great work. I will link this tool back in the blog post today if you don’t mind :)

Yeah, I am slowly removing myself from the internet, I don’t think that I follow anyone on any social network where ever I exist.

This is also a great way to backup tweets etc.

Of course I have multiple SQLite files with all the data, whenever I want to delete “more”, I just take a new Twitter dump.

Oh absolutely, just an experiment, I am really curious about doing things like video editing etc on the client side.

I did use this to delete a lot of old tweets though.

I am so excited about Prisma, I have been lucky enough to follow Graphcool for almost a year. I am sure with all the experience Graphcool team had hosting a Graphcool BaaS, prisma is a take on the harder problems in GraphQL ecosystem. From an initial look at the docs, I can see that prisma is language and database agnostic (with the help of connectors) and it has built in support for subscriptions (and more). I have a few questions about the service though:

1. GraphQL specification does not have the spec for subscriptions yet, moreover, even when added I am sure where and how to manage the state of subscription will be an implementation detail. How close is the Prisma implementation to what can end up in the spec? What does the subscription implementation detail look like, how is the state stored? I noticed from docs “Because Prisma is a standalone process, it can be scaled independently from your application layer and provide scalable subscriptions infrastructure.” Can we elaborate on where prisma sits in the stack?

2. Is prisma language agnostic? How easy it is to add new languages?

3. Is prisma DB agnostic. Can it be connected to an existing database or REST service.

I think questions 2,3 boil down to “How is prisma different from a GraphQL application server like graphql-yoga with GraphQL binding”

Awesome to see Graphcool team coming up with solutions to build a flexible and scalable GraphQL infrastructure.