Raw Notes

One thing I am noticing these days that all the successful vibe coders are developers, or at least, they understand the basics of programming. They just keep downplaying it by saying that they don't know how to code but still able to build successful apps.

Why?

Because it gives them more attention as people are more attracted towards something which requires less effort but does more impact.

I think, AI is going to speed up the innovation and it's going to bring more good to the world than more harm.

Think of AI as a tool, just a tool. For example, how inventions of "tools" like wheels or iron or other groundbreaking things completely changed the course of the then-future. Similarly, AI is a tool which is enabling people to do more people's work and that's indeed a great thing.

Yes, it's going to take some people's jobs but at the same time it's enabling people, who use AI, to work more, or rather, do more work with less efforts in less time. No?

What say?

Yes, vibe coding is really cool as it should be. But have you ever tried going a bit slow and also reading the docs and understanding what AI is writing? It's so satisfying, to be honest.

I was trying to implement auth to a Next.js project and Claude 3.7 Sonnet was really struggling with that. However, Google's Gemini 2.5 Pro did successfully implemented it and I was so happy. But I decided to re-do the authentication, but I was also going through the generated code this time; I was surprised that how easy the auth implementation was as compared to earlier when I was just asking AI to fix the issues again and again. If I had just followed the docs, it would have taken me a lot less time and I would have also felt more confident vibe coding.

Sometimes, it does feel that I don't need to read the code and let's just get it over with. But I am trying my best to, at least, just have a casual look at the code that AI is generating.

What say?

Like everything else, art evolves.

I keep seeing people not accepting the images or rather visuals that AI produces, and I get it because acceptance takes time.

But... if you think, were the current popular form of arts always like this? No, back then people did different things and now people are interested in different form of arts. And the transition phase is difficult, no doubts.

For example, initially, they draw on stones or leaves and then would they have easily accepted when someone first invented paper and starting drawing on the paper? No, absolutely not. But eventually, everyone moved to paper, right?

Similarly, people nowadays are making Ghibli arts using AI and then some people are calling it bad. I understand that, but eventually, this will be widely accepted and no one will say a thing.

This is how it works.

Viewing Branches

git branch                  # List local branches
git branch -a               # List all branches (local and remote)

Creating Branches

git branch <branch-name>    # Create a new branch (stays on current branch)
git checkout -b <branch-name>  # Create and switch to new branch in one command

Switching Branches

git checkout <branch-name>  # Switch to an existing branch

Merging Branches

git checkout main           # Switch to the target branch
git merge <branch-name>     # Merge specified branch into current branch

Deleting Branches

git branch -d <branch-name> # Delete a branch (safe - won't delete unmerged)
git branch -D <branch-name> # Force delete a branch (even if unmerged)

Remote Branch Operations

git push origin <branch-name>  # Push branch to remote
git pull origin <branch-name>  # Pull remote branch

Yes, coding using AI is harder than you imagine, especially when you try building complex apps.

For example, I tried building a web app using AI and it perfectly did everything I asked, but only at a point; after that it started hallucinating and the changes were not very good. And, in fact, I tried all major AI coding apps available in the market.

Here's what I think about AI coding tools:

• Lovable.dev: bad
• Bolt.new: good for simple projects
• v0.dev: really good, but highly opinionated
• cursor.com: the best (but sucks in UI)

Yes, I have tried only these as of now.

For the first time, I started experimenting with replying to people tweets wherever I see fit and I think it's a good idea, and it totally makes sense. I'd say that now I do understand people spamming replies under famous tweets. But I'm not talking about spamming, obviously.

I posted this simple reply under a tweet when it was just posted. And by the time of writing this post, which is about after 4 hours, it already has ~4,500 impressions. Most of the time, my tweets don't get this much impressions, but a reply did. Crazy!

But the question here is, like most of my tweets, like 99% of the time, aren't intended to "sell" something, this was also unintentional. I just liked showing my setup and people liked it.

I think, I would do this more often now. Not very aggressively, but yes.

Actually, I have recently set up a PostgreSQL database on a VPS for a new project I am working on.

And since this is the very project of this kind that I am working on, I am learning some basic SQL queries and will be taking a note of some of them below:

Common SQL queries

Here are the most essential PostgreSQL queries you'll likely use frequently:

1. Select all records:

   SELECT * FROM sample_table;

2. Create a new table:

   CREATE TABLE sample_table (id SERIAL PRIMARY KEY, name VARCHAR(255), created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP);

3. Add a new column:

   ALTER TABLE sample_table ADD COLUMN status VARCHAR(50);

4. Change column data type:

   ALTER TABLE sample_table ALTER COLUMN rating TYPE FLOAT;

5. Update specific rows:

   UPDATE sample_table SET status = 'Active' WHERE status = 'Working';

6. Delete specific rows:

   DELETE FROM sample_table WHERE created_at < '2023-01-01';

7. Insert new records:

   INSERT INTO sample_table (name, status) VALUES ('Example', 'New');

8. Rename a column:

   ALTER TABLE sample_table RENAME COLUMN old_name TO new_name;

9. Drop a column:

   ALTER TABLE sample_table DROP COLUMN unused_column;

10. Remove all rows from the table while preserving the structure

    TRUNCATE TABLE sample_table RESTART IDENTITY;

I just found out that Obsidian, the notetaking app, is not open source. I do not have a problem as I am already using tons of apps that are not open source, but to be honest, I always thought that Obsidian was a open source app.

I found out from this tweet, where the OP also thought it was open source until he found out.

I have used Obsidian for a long time, not so much recently, but still have installed on my computer and occasionally use it. As of now, I have started using the VS Code directly for taking plain-text notes and it's working fine for me.

Actually, the idea here is, I want to take notes which are:

• no proprietary format
• no vendor lock-in
• should be accessible by any apps
• can be opened for decades to come
• should be full-text searchable
• should take less disk space
• freedom to backup at multiple locations

Yes, I would definitely want these features from my notetaking app. And my current VS Code setup is giving me the flexibility to do all that.

So... I am all set.

Came across this tweet on X earlier this morning where they are discussing about the closing of a database provider company called Fauna. It has recently announced that it will be closing down in the next few months.

I am super glad about my choice of recently setting up a PostgreSQL database on a VPS server via Coolify. I can also automate regular backups and there are multiple benefits:

• I am not platform locked
• It would cost me a lot less than, say Supabase

Actually, I have always been very hesitant about using a platform where there is the slightest possibility of me getting locked out due to any reason – like proprietary data format or risk of getting banned/blocked and not being able to access the data.

Yes, I did not settle with such note-takings apps as well. Currently, I take notes in the plain .md format which can be accessible using any tool, and backup those notes to multiple places securely as well.

I saw another tweet about the same, and feeling good about my choices lately.

I got to know about the amazing hackathon that bolt.new is organizing via this tweet and I am all excited for it. Have already applied for the hackathon, but currently not very sure what and how I will be building it.

However, it's a good opportunity to push myself to do something under a given timeline while also standing a chance to win something. I know, there would also be "real" developers applying for the hackathon but I think, it's more about the idea than the code. Currently, not sure what we have to build so I am waiting on their official website to be updated with latest info so that I can think of an idea accordingly.

Super excited for this. LFG!

I am fascinated by BearBlog and tempted to build a similar but a bit different (and better?) blog platform. And Karpathy just made it an even stronger case with this tweet.

Someone asked and ideal features of a blog would be, according to him:

• a WYSIWYG markdown++ editor
• support for math, code, images, etc. (Obsidian style)
• basic features like SEO, feed, newsletter, custom domains, analytics, media, etc.
• discovery feature (much like Bear blog style)
• easy to leave the platform with all data in simple formats

"I mean... basically I want Bear but richer authoring interface that looks a lot more similar to Obsidian, instead of just a legacy plain textbox."

Yes, that's his last line.

I am sure that I am very tempted to and can actually build such a platform but not very sure about the tech stacks to make the code maintainable. Because if I start, it should be a long-term project that keeps running for years to come, maybe even till I live or maybe even after that.

I did take a look at the Bear Architecture and it's good. So... here, only my limited knowledge of databases, Django, maintainable code, etc. is holding me back.

Absolutely nothing else.

I was thinking about how much of my time screen.studio is saving for editing videos. I think, it saves me at least 30–60 minutes of time for each video that I put on YouTube – my workflow has become a lot faster because of this one app.

When I record videos, it saves a lot of time in editing:

1. automatically creates the zooms highlighting important areas
2. automatically puts and records my camera feed in the video
3. splitting and removing certain sections is much faster
4. now, i can also blur/highlight certain parts in the video

And much more.

Screen.studio is one of those purchases that save you a lot of time and basically offer much much more value than the money you spent on it. Super happy for it, and I don't think I will be needing a separate setup for videos anytime soon.

I always say that the world is full of coincidences, we control a very few things (or rather even nothing), and all the things around us are so random.

I am watching this podcast where Paras Chopra is talking about randomness, and I became so happy after listening to the argument because I also think the similar way. While we like to think that we're in control of our lives, no, it's not the truth – things are so random that you don't even know what's going to happen tomorrow or even in the next minute.

It's chaos everywhere.

And that brings me to this post of mine where I've discussed about "uncertainty" and how everything revolves around that only. You know, I started playing badminton and got my leg fractured only 3-day after joining the court. Clearly, I wouldn't have predicted it.

But what do you do with this information that everything is random?

Nothing, really. Just know that not everything is in your control, so if you be successful in something, happy; but if you do not meet success, do not be too broken. Because you did not have all the cards.

Yes, hyping works in marketing, you have hype a little that your product works great, solves real problems, and is getting good response from the people. I understand that; but what I don't understand is people, at least on X, hyping every single thing about AI.

It's irritating.

You might have also come across these folks who unnecessarily try to hype every single of their tweets. I mean, hyping a good strategy to stand out and make a bold claim, but when you hype everything, you're slowing decreasing the value of your next tweets/posts.

People will start getting bored knowing that all your posts are hype, and then your social account will die a slow death.

Don't do it.

China is not holding back!

It's dropping mind-blowing developments in AI one-by-one. It was only a few weeks ago when DeepSeek was released and now they have released this AI which has the capabilities of Deep Research + Operator + Claude Computer combined. I mean, just see the demo by yourself.

Yes, it's a bit slow as of now (claim a few people who already have the access), but hey, it's just the start and I'm sure that it will become even better.

Excited for this!

I don't know why some people are worried or rather pissed about Pieter Levels building a simple flight simulator game like this and making money out of this. Obviously, it's not about how good the game is (one person was complaining about this, I don't remember who now), it's about "reach", it's about "distribution", about "placement". Advertisers are paying because so many people are trying this "mediocre" game and it's going to give them eyeballs. Simple.

I love it, have played the game, and it's fun despite being very simple.

I also noticed that this has given a lot of other people the motivation to build something like this and, surprisingly, those few people are also making money out of this:

Very cool, right?

I expect, there will be a lot more this kinds of game would be coming out in the near future which can be played directly in the browser.

ffmpeg is great!

If you want to compress a video without visible loss in the quality, here's the command:

ffmpeg -i input.mp4 -vcodec libx264 -crf 28 output.mp4

And if you want to speed the video up by, say, 1.33x, you can just run the below command:

ffmpeg -i input.mp4 -filter:v "setpts=0.75*PTS" -filter:a "atempo=1.33333" -r 60 output.mp4

And you can also combine both above commands into one, just run the below command and your video will be compressed as well as become 1.33x faster in seconds:

ffmpeg -i input.mov -vcodec libx264 -crf 28 -filter:v "setpts=0.75*PTS" -filter:a "atempo=1.33333" -r 60 output.mp4

Normalizes the audio of the video with a single command that I record using Screen Studio + Krisp.

ffmpeg -i input.mp4 -c:v copy -c:a aac -af "loudnorm=I=-17:TP=-1.5:LRA=11" output.mp4

Make sure to cd into the folder where your video file is, and replace input.mp4 with the actual video filename you have.

I thought, I liked the having serif fonts on my personal website. But I was wrong, recently, I switched back to Inter and I think the site looks so much cleaner and modern.

For now, I would be using this Inter font, until I find a good one.

Actually, I found a good one called "Helvetica Neue" (inspiration from simonwillison.net), briefly googled but couldn't find anything solid on from where and how to use it. Will look it up again.

Came across this screenshot that someone shared, it was a response from the newly released OpenAI's GPT-4.5, and I immediately related to it. It just felt so true.

Humans never genuinely pursue happiness; they only pursue relief from uncertainty. Happiness emerges momentarily as a byproduct whenever uncertainty briefly disappears.

I screenshotted the screenshot that you're seeing above.

How accurate!

I have been using Placehold.co for a long time, and still didn't know that you can also add custom text, colors, fonts, etc. directly from the URL. For example, take a look at the below URL:

https://placehold.co/1200x630@2x/f9fafb/2563eb.png?font=playfair-display&text=DeepakNess

When you open it, you will see a retina 1200x630 image with "DeepakNess" written over it using the Playfair Display font.

How awesome!

Yes, truly.

I get bored working on old projects but working on new projects is always fun. I can be occupied for days without getting tired, as I am already doing the same with a current ongoing project (Search To Be). I'm good and settled for at least a month with this project.

Been creating new posts and pages, collecting useful resources, and whatnot.

Let's see how much time before I actually get bored with this one.

I came across this tweet by @levelsio and then did a Google Search to find which AI SEO related terms are actually understood by Google for what they actually is.

And it turns out:

1. AI SEO – 19.9k monthly searches
2. LLMO – 1.6k monthly searches
3. LLM SEO – 230 monthly searches

All searches are according to SEMrush.

Apart from the above terms, others are not properly understood by Google. For example:

• GEO is understood as geography-related
• LEO as zodiac sign
• AIO as all-in-one cooler (have no idea what it is)
• AEO and GSO are completely different as well

So... I think AI SEO and LLMO (Large Language Model Optimization) is going to win for now.

How does the AI-based search engines work?

They still rely on "content" and I think whoever has the most content in the AI-powered search era, is going to win. After all, blog posts are not outdated at all, they just became sexy once again.

I think, any new or old businesses should keep creating more and more "original" content.