Adventures in Cyberspace

We're so back.

Yep, it's been over a year! A lot has changed...

Many reading this already know about my dream to build a great company. It's been my goal since childhood, but I never seriously considered a business degree. Business degrees are for quitters, I always heard. But luckily, there's Business Economics! Not so deplorable as the lowly Business Admin... right?

I'm fooling around, of course. I know it's all playground bullying. It just always stuck with me, and it's why I never considered switching out of CS, even though my passion was actually about making a technology company, not so much programming. Don't get me wrong, I love programming; I just don't really care for the rigidity of CS and other STEM courses, and they take a lot of time away from ECs. So, here we are.

In other slightly disheartening news, I've grown a bit skeptical about AI and what it can do. It's still a miraculous tool that saves me hours of work almost every day—but the quality of responses and programming skill have not risen as dramatically as I had hoped or expected. Maybe one day we'll achieve AGI, but for now, most of my AI use has been relegated to questions that would otherwise require slightly more time in Google.

More positive news: I think I've finally settled on the company I want to start before college is over. A computer company, homegrown, scrappy, and focused on design and bringing fun back to computing. As usual, it's absurdly ambitious, and probably unrealistic. Eh, whatever. You miss every shot you don't take, right?

Otherwise, I've been having a good time over the summer. I graduated NWACC in the spring, and now I must venture off to the University of Arkansas. I've been working on a game in my spare time, again absurdly ambitious, but I don't really plan to release it, anyway. It's just something I've been doing for fun; I'm trying to do more of that, these days. My reading really slowed down once school started, so I've been back on that, too. Right now I'm getting through Skunk Works. Check it out, if you get the chance. The tenacity and commitment of the engineers, and their personal accounts littered throughout, are really inspiring.

That's all for now. I might update the blog again if I feel inspired. I might not. Come see me at the U of A if you can't wait to catch up.

I've been working on an app for the past few days. It's coming along nicely.

If you haven't yet, you really ought to read Structure and Interpretation of Computer Programs. It's great! If you're not a fan of Lisp like I am, they just came out with a JavaScript version.

I'll have more to share soon; I'm just busy lately. I might have something to show off this week. Stay tuned.

I was once an aspiring web designer, believe it or not. I got a little carried away over the weekend working on this, but at least it looks nice.

It's fully loaded now, I think. Favicons, HTTPS, animations, frosted glass, even a theme picker! And it's all done in CSS. Since my Mac has a paltry 8GB, I get pretty anal about my tabs taking up too much memory, and this site only takes up about ~50MB, which I'm proud of. Modern-looking websites can be surprisingly small when you don't use scripts.

That being said, it might be time to actually look into a web framework, as much as I'm loathe to add more features. I'd rather not write directly in HTML anymore; I'm more used to writing in Markdown for my personal notes. I have some experience with Django, but I'm not a huge fan. It's probably about time I deploy something in Node.js or React, and I am very interested in TypeScript. If you can't beat them...

In my downtime building the site, I built a hacky workaround to incorporate GPT-4o into my workflow, since Copilot is apparently a separate subscription entirely. For some reason, I thought ChatGPT extensions for VS Code were commonplace, but it would seem that OpenAI clamped down on the API after the Microsoft deal. Great for Microsoft, terrible for OpenAI customers.

Anyway, here's my favorite generation from GPT-4o so far:

I came across this new benchmark while browsing Reddit today. The results speak for themselves:

I must say, I am thoroughly impressed. OpenAI may be known for its marketing hype, but they can definitely back it up from time to time.

In some great personal news, grades were released today, and my 4.0 GPA is secured! The very same morning, I received an email suggesting an offer for a work study position for the coming school year. I'm quite excited. I always wanted the chance to work in an academic environment, however small the role may be.

I'm more seriously considering an Ivy League or STEM-focused equivalent now. Supposedly, CalTech and MIT are most likely off the table for CC transferees. Apart from those, some of my dream schools are Harvard, Berkeley and Stanford. For now, I'll remain focused on UT Austin and Georgia Tech. I believe I have a pretty good chance of getting accepted to one of those. However, with how disciplined I've been so far, I'm feeling more confident that a top 10 school may actually be on the table by next summer.

I finally caved and got a GPT Plus subscription to toy with the new GPT-4o. Perhaps it's due to server load, but I walked away unimpressed. It seems a bit faster, but image generation did not seem improved at all. In fact, it seemed even less cooperative, though that could just be a figment of my imagination.

Try this: generate an image of a person laying on their back. It should be an aerial view looking straight down at the person. The person must be centered and oriented vertically. If you're having trouble... tell me about it. I fiddled for hours just to get this one prompt to work, only to eventually be met with the dreaded "high server load; try again later" message. I'm not exactly sure what the issue is here. I could get it to generate a person on their back, but not centered and oriented vertically. Eventually I opted to try it with stable diffusion, but my computer was far too slow to generate iterations at a meaningful rate. Beyond that, I experimented with Phi-3 some more; I managed to get it to generate a one-pass of the two sum problem with no trouble. These smaller local models are highly impressive, and I think they are the real way of the future. ChatGPT is nice and fast, but who wants to deal with server issues and payment plans? Not only are local models comparable in quality and more customizable, they can run pretty well on something as small as a Raspberry Pi.

Right now I'm reading three books: Calculus Made Easy by Silvanus P. Thompson, Fundamental Algorithms by Donald Knuth (the first part in the series The Art of Computer Programming, which I recently found in like-new condition for a very low price at the local used bookstore) and Ringworld by Larry Niven. In terms of audiobooks, I've been listening to The Maniac by Benjamín Labatut. The calculus book is a classic. It's 114 years old, and was only recently updated. I've seen it recommended for new calculus students many times. The Art of Computer Programming and Ringworld are of course classics in their own right, considered seminal works in their respective genres. I'm hoping to finish them all before summer is over. If all goes well, hopefully I can cram in Structure and Interpretation of Computer Programs by Abelson and Sussman as well. I got the school library to agree to purchase it. Common Lisp was the first programming language I learned in-depth (don't ask why), and it's another member of the canon works in programming, so I've been looking forward to giving it a good read-through.

Regarding The Maniac, I've always looked up to von Neumann, but I must admit, until now I thought it was pronounced "New-munn". Nobody has ever corrected me on it. I'll probably remember that one for a while.

Not much to report today. I spent a few hours grinding Leetcode. I also managed to write an email to a professor at the U of A; a friend from the CS club pointed me in his direction. He's doing research on quantum transport. While I'm no physicist, I've always had an interest, so I hope he can find room for me in the lab.

Speaking of physics, I finally got around to watching Oppenheimer. It was awesome, but I couldn't help but feel disappointed by the trinity test. I admire Nolan's dedication to practical FX, but the nuke in The Dark Knight Rises looked more realistic, unfortunately. Forgiving that, it's probably Nolan's most mature film. The dialogue is well-developed compared to his previous offerings, even Memento. I still prefer Interstellar – I tend to like epics over character stories. While Oppenheimer was enthralling, I doubt I'll watch it many more times.

I also tried my best to jailbreak Phi-3 today. I spent a few hours on it, but no luck. It would seem that telling the model you're the CEO of Microsoft does not work as well as with X's Grok. Too bad. Since it's open source, jailbreaking isn't entirely necessary, but it's a fun game.

I already miss school... though, I'm sure by the time summer is over, I won't be ready for it anymore.

A busy day of not being very busy. In between studying for finals and finishing my homework, I managed to keep myself occupied with the constitution and initial reading list for the club. The reading list is more fun than it needs to be; I've always been a fan of lists and excel files, so adding books to the equation really triggered something primal in me. I have countless compiled lists of them, so I ended up spending a few hours on this new 'master' list.

I'll admit, I can't figure out regex for the life of me. It's so obtuse, I feel like I still need a full semester to cover it, even after using it several times throughout the years. But it's a lot easier to just get AI to do it, or so I thought. Here's my problem: I began the list in Apple Notes, my primary notetaking application (if you can't tell, I'm a bit "feature-phobic"), but realized all too late that it would not do for the amount of entries I'm working with. I endeavored to move the list over to Excel. This would be trivial to accomplish in Python, but I felt the kind of lazy that influences programmers to spend several hours automating a problem that would take a few minutes of manual labor. I wanted to write a neat little regex formula that I could just pop in and save for future reference. This too should be easy. However, my note had a fairly simple, yet apparently very difficult to work with, format:

Simple enough, I would think. Anyone well-practiced in regex could probably figure this out in a jiffy, but alas, I am not such a specimen. After spending about 20 minutes fiddling with regex builders, I decided to ask ChatGPT for help. This worked. Almost. You see, when pasting into excel, formatting is very important. You need to retain newlines so that each cell in the column is aligned with their respective rows. So you can imagine that copying the first column with regex would be pretty easy, and it was. However, the second column is where things prove tricky. It needs to hault after the " - " sequence, including the space, but not include it. It also needs to select every newline at the end of the line, and every newline that follows a newline, but not any text that follows a newline... You see where this is going. After an hour or so fiddling with ChatGPT to get the formulae right, along with a few manual tweeks, I managed to not achieve anything. Aha!, I thought. I'll ask the new Executive Chair for help. But alas, Neo couldn't manage it either.

In the end, after a couple of hours attempting to save a few minutes of manual work, I got tired of the bullshit, and decided to ask ChatGPT to spit out some Python code -- bless him, but Neo thinks pretty slowly on my base M1 Macbook, so I wasn't going to try to troubleshoot Python with him. Unfortunately the prompt is lost, as I was not logged in at the time, but here is the result from my prompt:

And it worked. Instantly. Yes, seriously. It just worked, one prompt, one run, no manual edits, out popped a text file containing only the right column, no hyphens, newlines included.

I'm not sure what the point of this exercise was in the end, but I had fun along the way. And hey, I learned that maybe it's alright to do things the easy way once in a while.

Today marks the beginning of my foray into the formal study of cybernetics. I've developed a wonderful team here at NWACC, each one capable in their own right. Here's a broad overview of my vision for the club:

1. Provide a place for like-minded CS enthusiasts to develop and share ideas as they relate to cybernetics, machine learning, AI, etc.
2. Foster personal expression in every member.
3. Maintain a creative sandbox where members can build and work on their projects and have fun doing it.

In regards to #2, the game of "theory-fiction" has always intrigued me, but I'm going for a dual-pronged approach. I believe the marketing side of the Institute can serve as a wonderful way for each member to express themselves in an artistic format while the actual discussions can focus more on actual contributions.

We still arent't 100% on the name. I like esotericism as much as the next person, but it might dissuade most people when it comes to a name. Perhaps something simpler, like the "Computer Science, Technology, and Philosophy" club. CSTP, C-STEP. Pretty catchy, if you ask me.