sarmadgulzar

I wish more people understood that prototype != product

Exactly, I absolutely agree that AI supercharges seniors with judgment and conviction. They’re no longer constrained by how fast they can type.

But this is about what comes next i.e. seniors of tomorrow

If AI gets so good at software that vibe coding is the new norm and as Elon Musk says that AI will generate the machine code directly without any intermediate compilation or interpretation then we wouldn’t need seniors or juniors but I don’t think that’s coming any time soon, if at all!

Sure you can do it but I think you’ll agree with what I said about ownership. Whenever I ask AI to audit my codebase, it always plays it safe and gives me a few options to choose from. It just won’t give me a concrete answer unless I insist it to give only one but then I wouldn’t trust that answer that I basically forced out of it.

Agreed. I’m not insinuating that colleges should teach all of these extremely domain-specific things. Most of the stuff would indeed be unnecessary for most students depending on what they do in the future. But that is my original point that you learn this on the job while making a mistake and being corrected by a senior mentor. AI wouldn’t call this out by itself unless you give it a very specific prompt which you also can’t write if you don’t know that these kinds of problems happen in the first place.

Precisely my point. But unfortunately companies are rushing to get the features out. No cares if you’re learning and augmenting your own understanding or not.

> Shouldn't this be a pretty fundamental part of a degree? If you're in school right now for a CS or SWE related degree, I would imagine you're learning how code actually works, how the math actually works, etc.

I’m sorry to say but colleges don’t teach the kinda stuff you’re gonna use day-to-day. They’ll focus on architecture, data structures, algorithms, databases, OS internals theory that only a minuscule number of systems engineers would get to work on. They won’t teach you that in your /login endpoint, if a user is not found, you should verify the password against a pre-computed dummy hash so the response delay matches a real user account workflow to avoid timing attacks. You only learn this on the job under the supervision of a senior mentor.

Stakeholders are not technical enough to understand that so CEOs won’t do it. We are already seeing it play out. Entry-level hiring has been consistently down ever since AI got mainstream.

This all sounds good on paper but I have an honest question: How do you develop judgement as a junior or mid level engineer to critique AI if you yourself never learned that skill by making mistakes in the first place?

I’m not saying that all code should be typed by hand in 2026 but there are certain subtle things you learn only when you get into nitty gritty details especially related to security.

Also today’s AI is notoriously bad at ownership. When you ask it to give you a concrete answer, it will still give you options with pros & cons of each so that ultimately you own the decision and not it. So how do you decide between the two (or more) when you never learned to do it yourself?

> AI has, multiple times already over multiple different era's definitions of what counts as "AI", been used to invent new things that humans have not been able to.

Can you give a few specific examples?

> They can be superhuman one moment and amateurish-to-childlike in the next.

Even the smartest humans are like that. No one is best at everything. Which brings me to another question: Is a know-it-all ASI actually possible? If so what's the bottleneck? The current approach, the data or the compute?

PS: These questions are kinda pointless. I understand that we don't know how to answer them yet hence I'm not expecting them to be answered. Just your general thoughts.

Writing the blog post took longer than the million line Rust rewrite. That is all someone with a few brain cells needs to understand what’s going on here.

Are you from Pakistan? I have an Alchitry Cu FPGA board for my personal use if you want to borrow it for implementation. Awesome project btw!

GitHub: https://github.com/sarmadgulzar/xiao-ble-embassy

RP Zero 2W is $15 and Uno Q is $44

Thank goodness I switched to nrf + embassy before this happened

I didn't say perfect, I said closest to perfect. Regarding the implicit return types, it's all a matter of taste. I think they're very clean, but Rust is not forcing you to be implicit—you can be explicit if you like. e.g., if a function's return type is String, then whether you write the implicit "Hello".into() or the explicit "Hello".to_string() or String::from("Hello") is entirely up to you, and Rust will not complain.

I know I'm biased, but Rust is the closest thing we have to a perfect programming language. Is the borrow checker a pain in the ass? Yeah. But is it necessary? Absolutely. Imagine writing the same buggy program in C, deploying it, and then it blows up at runtime—you still have to fix it, right? A bug is a bug, and it needs fixing. The difference is Rust forces you to deal with it before you even get a binary, while with C you might get a 3 a.m. wake-up call trying to figure out what went wrong. So it’s not that Rust is harder, it’s just different. It takes a paradigm shift in how we think about writing safe and secure code. Change is uncomfortable in general for humans and that paradigm shift is precisely why most (I hope not) people feel this way about Rust.

I see that you're using gemma3n which is a 4B parameter model and utilizes around 3GB RAM. How do you handle loading/offloading the model into the RAM? Or is it always in the memory as long as the app is running?

Can relate. I've also shifted towards generating small snippets of code using LLMs, giving them a glance, and asking to write unit tests for them. And then I review the unit tests carefully. But integrating the snippets together into the bigger system, I always do that myself. LLMs can do it sometimes but when it becomes big enough that it can't fit into the context window, then it's a real issue because now LLMs doesn't know what's going on and neither do you. So, I'll advise you to use LLMs to generate tedious bits of code but you must have the overall architecture committed into your memory as well so that when AI messes up, at least you have some clue about how to fix it.

So, this can be solved in at least one and at most two attempts.

I agree with most of what you said. For me though, it's really hard to go to YouTube without wasting at least an hour. That's why I built this app. So far, it's working great for me. I've added my favorite channels into different lists based on categories and set up weekly reminders for all of them. For example, every Friday evening at 7 PM, I get an email about the latest videos from Veritasium, 3Blue1Brown, Mark Rober, Ben Eater, Steve Mould, SmarterEveryDay, and Tom Scott. This way, I don't feel the urge to go to YouTube and lose track of time.

True. But Watchlist is not just a video aggregator, it’s got many more features like customizable alerts and progress indicators etc.

I’m currently working on a feature to let AI determine if a video should be added to a list based on what it’s about.

Tried and failed. When you turn off history, the algorithm will actually suggest even more enticing videos.

Generating playlists is one of the things that Watchlist does. Here are a few more: 1) You can actually set very flexible notification schedules and use a medium of your choice (email, webpush) to receive the notifications about new/unwatched videos in a list.

2) You can keep track of your progress by marking the videos in a list Watched/Unwatched. This is good if you’re taking a course on YouTube.

A few things I am building in near future if all goes well: 1) Adding the “tags” functionality which will suggest similar channels to users based on the kind of videos & channels they’ve added.

2) A video summary feature which will send a short summary of the video to you.

Thanks for sharing this. This tool is a very small subset of what I am trying to do with Watchlist.

1) In the current version, you can actually set very flexible notification schedules and use a medium of your choice (email, webpush) to receive the notifications about new videos.

2) You can create lists by cloning YouTube’s playlists and then keep track of your progress by marking them Watched/Unwatched. This is good if you’re taking a course on YouTube.

3) I am working on adding the “tags” functionality which will suggest similar channels to users based on the kind of videos & channels they’ve added.

4) I will also be adding a video summary feature which will send a short summary of the video to you.

I have many more features in mind, just need to get some initial validation of the idea.

Can you please elaborate a little?

A quick demo of how Watchlist works: https://youtu.be/NqgZelLWsPs

Love the name. Fits well.