Ruksi Korpisara

@JuhaKiili where can I play it 🤩

Day #1 of making a game It’s not really day #1. I started noodling around with this two weeks ago, but let’s say this is the first official day of dev. The game is going to be a city builder. The “frontend” is Love2D, and the “backend” (sim) runs in a C library I’m writing from scratch. I want to support a LOT of agents (think Songs of Syx), hence the custom library for fast, multi-threaded simulation.

@dhh @jdxcode mise is also great for tool installs; I work with more than ten different k8s clusters and mise solves that "kubectl should have one minor version difference of your cluster" issue just cd to a different directory (also configures right kubeconf via mise env vars)

The sweet spot for Docker in development is to let it run your accessories, like database, Redis, ElasticSearch, but stick to native installations of Ruby/JavaScript/whatever using mise. No "works on my machine" version drift, little fuss, great performance 👌

@JuhaKiili wipe or poop? no, I read that wrong 🫢

In game jams, people come together to create a video game in 48 hours. I was supposed to join the Global Game Jam, but got sick the day before. This year's theme was "bubble," so I made a bubble puzzle game anyway! See if you can beat all 7 levels: tinyurl.com/swipe-or-pop

@DegenRolf so my brain uses about as much power as a light bulb but is capable of a lot more complex tasks... and here I am using it to scroll social media

The human brain runs continuously at a modest ~17 watts of power and only increases its energy consumption by 5% above baseline when it is actively thinking. The brain is organized to minimize energy consumption while maximizing computation. This means that, while the brain consumes the largest proportion of energy in the body, it is remarkably energy-efficient considering its computational power. Here, we review the metabolic costs of cognition, that is, the how glucose metabolism sustains brain functions including core homeostasis, memory consolidation, repair and the execution of specific cognitive tasks. Although it accounts for only 2% of body weight, the human brain accounts for 20% of its resting metabolism, more than tenfold the amount expected based on its weight.  However, the brain nonetheless runs continuously on only ~17 watts of power. By comparison, a large high-performance computing cluster uses up to six orders of magnitude more power, operating at ~2 megawatts. From this perspective, our brains are remarkably energy efficient relative to their computational depth and agility. Relatively simple unimodal tasks (such as visual perception and visuospatial processing) are less costly that complex multidomain tasks (such as social cognition and emotion). Although these observations regarding energy utilization are convergent, an important caveat is that engagement in explicit goal-directed behaviour and cognition is only associated with a relatively small increase (~5%) in glucose consumption [compared to] the ongoing costs of resting neural activity and homeostasis. This is consistent with evidence from neurophysiology that the majority of the energy budget of the brain is spent in maintaining core homeostatic processes plus neural activities that are not tied to immediate task demands such as memory consolidation.

@JuhaKiili it certainly boosts output of individual devs! easier to juggle multiple projects and context switch around creating and managing context files like PROMPT.txt etc. also help you to stay on top all the important bits and pieces of each project 😅

Here are three hypotheses based on my personal experience using LLMs: 1. Writing code gets the largest productivity boost. 2. Attending meetings gets the smallest productivity boost. 3. The coding-to-meeting ratio is a simplified proxy for the total LLM boost. From my experience, the larger the organization, the more communication is required, which generally means more meetings. Therefore, I believe that the smaller the organization, the bigger the boost from LLMs. Of course, there is plenty of gray area in between, and every organization is unique.

@fidexcode TL:DR: it's not a niche, it's the mainstream web dev offers lots of job opportunities as all orgs need web sites, apps and services and most projects have UI so you can take dips to artsy designer side of things plus, the web is always evolving offering constant growth

Why does everyone want to learn web development?

@DevKhan03 GNU Stow

What's the first thing you always install on a new computer?

@shavkatrakhmano @Drealtemiteee_ @andreiacribeir exactly! would remember such an occasion

As a previous tech recruiter: No one cares about your Github commits during the hiring process

@RaghavT_ @andreiacribeir I don't personally mind them, some might see them as unprofessional

@ruk_si @andreiacribeir How do you spot the fake history? Also, how do you feel about those developers who write "Hello", "Hey", or some other text using their commit history like this 👇🏻?

@Drealtemiteee_ @andreiacribeir you get a feel for it; too predictable patterns are the easiest to spot inability to explain context on commit spikes is another that might be a giveaway if I had a big 70 commit day or rebase, I remember it but nothing is foolproof, ofc

@ruk_si @andreiacribeir How do you spot the fake history?

@jxnlco scope your uncertainty to where it matters most tips imo missing things about tight evaluation boundaries around prob components that let you fail fast + iterate quicker but I do agree to focus on "the end goal" behind the metrics you measure

@thdxr usernames are a solution for privacy when the platform has public content 1. people don't want to show their email on the site 2. you don't want to appear as "anon123" using the first part of email is an OK default, but some people have silly emails so it has to be editable

i am begging you stop asking for username - if i need multiple accounts for work / personal / etc this is annoying af

@marktenenholtz and just like the first time you pretend you know exactly what you are doing but really just make things way more complicated than they needed to be

The engineering equivalent of losing your virginity is running into the Second System Effect

... but I'll cap it for now, before it gets way too geeky 😅

@javilopen some creations grow beyond their creators' design there are parallels to be made with AI vs. Lord of the Rings and the creation of Ainur like how Melkor was created as one of the most powerful Ainur but his actions led to unintended destruction, even tho that wasn't Eru's plan

“God didn't create humans. Humans will create god.” The question now is, will it be a god.. - Benevolent / Value-aligned - Malevolent / Unaligned - Indifferent Beware: a tremendously powerful indifferent entity can be as dangerous to us as a malevolent one. Just as you are to an ant when you unknowingly step on it. Your goal wasn’t to harm it... it just happened to be there. Similarly, an entity with powers we might perceive as godlike could decide that its next project, or a legitimate goal, is to build a Dyson sphere around the sun to harness its energy for some endeavor... thereby eliminating all life on Earth almost unintentionally. Think about it.

@ChShersh but if you come back to zero notifications that means you've already been replaced 😥

@ChShersh and if you come back to 147 unread notifications marked "urgent", you're practically CTO material at this point

If you return from a 2-week vacation and you don’t have JIRAs still assigned to you from before you left, your job is not secure enough. Take notes.

@burkov while I 100% agree, defining problems clearly is itself a crucial problem-solving skill maybe prompt engineering is making us better at breaking down complex problems into manageable pieces so I'm cautiously optimistic!

Google has made us incapable of memorizing complex things by reducing memorization to search queries. LLMs will make us incapable of solving complex problems by reducing problem-solving skills to writing prompts.

@burkov humans are more reliable at being unreliable

Working with people: out of 10 people, you get 8 times work well done and 2 times you get crap. Working with agentic AI: out of 10 jobs, you get 10 jobs done at 20% crap ratio spread all over the place nicely camouflaged.