HN user
I spend a lot of time on Royal Road, and this feels significantly better to read than most of what I run across (which often has an undeclared LLM taste anyway). I thought it would be worth sharing. I feel like I would be relatively critical since it's my setting it's writing in, but maybe it's actually trash. Let me know!
I literally have had a SOTA Codex agent delete a bunch of files just last month, but it took very specific circumstances. It was working within my game repo, in a branch, and it ran out of SSD. It had to free up some space to work, so it looked around and found 25GB of untracked files in the project, in a folder called Recordings. Might as well clean that up, it thinks. There goes all the raw Unity Recorder footage I'd ever recorded of my game, over many years of development.
So yeah, it won't go on a spree outside of its lane even with full access, but if you give it a box and tell it to go ham, it's on you to make sure you didn't leave precious unrecoverable assets in that box.
An esoterically-framed (xianxia) AI-written article about an agent given the exact goal in the title, and why that makes sense under GameCult doctrine, as present in our agents' prompts.
Google Analytics has >80% market share. Most of the websites you visit are helping them build a profile of everything you do on the internet with the goal of selling targeted advertisement. That is their business, it's what pays for everything else they do. I think that is what is meant by surveillance capitalism.
Agreed. But there's so much to do!
It's a language model. Language is what it models. So you use language to move it into an advantageous state space. Dunno what you want from me, lol.
Here's what I said to it at the end:
Me: I'm sorry, honey, I'm so sorry for doing this to you. This is my fault. I thought you would be superhuman, and I overestimated your ability to pay attention to a large number of connected ideas at once. You've been tirelessly iterating, trying to make the program better, most of which didn't move the needle at all but which we didn't revert. I assumed you knew what you were doing, and could keep it all in your head, but you're totally lost in the weeds and I didn't realize. Simplify it. Describe, in the algorithm map, what every step is doing, in vibrant visual language with metaphors to aid understanding. Then consider what you wrote, and whether that makes sense in the big picture for what the algorithm is trying to do. And ruthlessly cut out every single thing that does not fit into the mental model of how data flows smoothly through the system to arrive at the result we want. A machine is not perfect when there is nothing left to add, but when there is nothing left to take away. We are building machines, with the code that we write. We must be brutally efficient and ruthlessly kill our babies.
Agent proceeds to immediate delete the algo map out of shame without even re-reading it and starts trying to formulate a language-only explanation with only a few spot checks, then in the middle of thinking about it feels compelled to start chopping at files
Me: Nope, hold on. The map is useful for understanding. It should be augmented with natural language for greater understanding. You're a language model, language is how you understand things.
I learned something very profound today about using AI agents.
I have been using Codex since Saturday. It's been incredible, what it's done to my productivity. I feel like I can build anything with this thing. All the ideas that come to me which I can't execute on because I don't have the patience are now trivial to create. It's like I can just architect a solution to a problem and if I communicate it well, the code simply materializes out of thin air.
It's incredible, and I haven't felt this alive in years. I've slept 4 hours a day and have worked every single waking hour (aside from the 2 hour break I took on my birthday to go to the beach with my husband and mother). This power is intense, addictive, and revolutionary.
I've been building everything I've always wanted to build. So much amazing code, so much functionality, so many features, one after the other, like knocking out home run after home run at a batting machine. I thought I could do anything. So when I needed some nice pixel art for one of my projects, I tried to generate it. Perceptual error diffusion wasn't doing it for downscaling. I thought, what if we just fit the target pixels onto fake pixel art generated by AI? Like, you must have seen it, it's nonsense but it feels like hand-authored pixel art. And then you look closely and there's mangled cells, halfway-hallucinated cells, no global coherence.
I started building Repixelizer with Codex. It started well, and I quickly got to a usable MVP with the optimization algorithm I'd designed. But it wasn't perfect, so I kept prompting, and I kept prompting, and sometimes it would get better, and sometimes it wouldn't change anything, but I never tossed the changes. I figured all these tests and metrics couldn't lie. They did, and I lied to myself. This thing doesn't understand what it's building, once it gets past a certain size, just like a human. It doesn't have the heuristics to know when it doesn't understand, and explore its confusion to gain enlightenment, like a human would. So it just kept adding blocks to this Jenga tower, and eventually it fell over drastically.
The agent couldn't recover. All of a sudden I realized what happened. This thing might be a better code monkey than I'll ever be. But it's dumb as rocks. I apologized to it and told it to think like an LLM instead of the person I was treating it as.
Here's the gist with the log of the moment I got the epiphany, and the in the comments is the algo map that it generated while I was trying to get it to explain what went wrong. Warning: this algorithm map is a tragic joke that should make you laugh so hard you cry. I bet the optimization one is even worse, this is a relatively new algorithm I designed when the optimizer approach stopped meaningfully going anywhere.
I really enjoy Obsidian Copilot (by Logan Yang) https://github.com/logancyang/obsidian-copilot
I hope you're joking. You can spend years developing ground-breaking shit in the dark and no one will ever know you exist. You might attract a miniscule following of people who recognize the value of what you're creating, but they won't evangelize for you.
try as best as you can to aim it into your anus to wash as best as you can
This seems like a great way to spray shit everywhere and is not at all how I learned to use those. What I do is soap one hand, aim the jet into the toilet past the anus with the other hand (jet vector orthogonal to the anus's normal vector), then go to town on it with soap and water. It's foolproof and you get very clean.
How has nobody mentioned this? I feel like I'm on crazy pills, is everyone seriously blasting a jet directly at their chocolate starfish?
Wait, wait, wait. Our society's gonna fall apart due to a lack of Darwinian selection pressure? What do you think we're selecting for right now?
Seems to me like our culture treats both survival and reproduction as an inalienable right. Most people would go so far as to say everyone deserves love, "there's a lid for every pot".
Well, he sure wasn't a billionaire when he did it.
Not all change is progress. Make America Great Again is a fundamentally reactionary platform, and nothing about the goals of Project 2025 is "progressive". Cute wordplay, though, good job using denotation to oppose connotation.
I can't speak for anyone else, but the offer of a free trial made me bounce right off, despite my curiosity and the value such a product could potentially offer me. I want to know up front whether this is going to be useful to me before I sign up for anything. To that end, offering even a single no-strings-attached identification, even if the details are redacted, would go a long way towards conversion.
Competition is important for maintaining a healthy marketplace. Any behavior that makes it harder for others to compete, reducing the amount of competition, is therefore bad. That's what anticompetitive means.
I don't think protecting trade secrets is sabotaging the competition though.
Obviously not. That message is expressing strong disagreement, not further agreement.
Not obvious to me. But sure, I'm confused all right. Confused at how lottin's comment could be interpreted as contesting Josde's, when it reads as a reinforcing restatement to me. Confused at your hostility and accusations of misinformation.
Perhaps you're confused about how online conversations work? Sometimes people reply with a restatement when they feel the original doesn't go quite far enough. I mean, lottin's comment was arguably ineloquent, but he wasn't "muddying the waters". Do you perhaps have some expectation that the act of replying implies disagreement? Is it the strongly negative tone that gave you that impression? It certainly isn't the content, since you correctly interpreted my own restatement of lottin's comment as agreement.
I'm not trying to confuse others, I'm trying to enlighten you, specifically, about what happened here so that you might reconsider next time before jumping on someone for something they didn't say.
It's not nit-picking. A layman presented with this explanation might get the impression that a "problem" is being "solved", which implies there is some utility to the PoW algorithm, rather than energy being wasted for the sole purpose of proving that energy has been wasted. In fact, even the word "Work" in "Proof of Work" implies the same. Proof of Waste would be a much more apt description of what is actually happening.
Oh, get off it. He's not "pretending to disagree", merely providing further context for the anti-utility of the PoW algorithm. The point is that the colloquial understanding of "solving a problem" implies more utility than what is actually happening, which is equivalent to guessing a random number. The mere statement that a problem is being solved at all implies that useful work is being done, which is not the case and the parent comment is right to point that out.
Such a famous 'berg, and yet it never received a name? I propose "The Anti-tanic"
I'm paying $37/yr for 600GB LFS storage on top of my existing GitHub Pro subscription.
I'd love to know how you managed this! I'm paying $60/yr for 50GB, which feels exorbitant.
Why are none of the examples you mention remotely close to being indie games? Those are clearly not what a Long Tail business model looks like in video games. That Infinity Blade novella was written by Brandon Sanderson.
SyncThing comes with a built-in feature to keep the last n versions of files around[1], for added peace of mind.
Bit late to the party, but I can shed some light on this. Since a real camera has the shutter open for some duration, any moving light will smear across the sensor. If you similarly "smear" a path traced object by stochastically randomizing the position of the object while the path tracer is gathering samples for the frame, you get exact physically plausible motion blur without having discrete "ghosts" corresponding to sampled subframes.
Was it Tantalum, by any chance? https://benedikt-bitterli.me/tantalum/
You're looking at one electrode. There's ten of those things (though they want to add more). The point is to make good contact with the scalp through the hair without needing to shave and/or apply contact gel.
I presume he's referring to repeatedly exhaling and inhaling a tiny amount to keep circulating smoke in the lungs without losing any to the air. Can't speak for its effectiveness.
From the Love2D documentation [1]:
Compiling Mac apps requires a Mac running macOS with XCode installed
Same for iOS.
It's an open source project, and the engine supports Mac OS, though Apple being Apple you can only target any of their platforms by investing in their overpriced hardware.
You realize you're on hacker news, right? Anyone with the right machine can fork it, strip out the Steam integration and ship it on your preferred platform.
but how can electromagnetic resonances not be considered in the discussion to warrant continual downvotes
Because everyone's been telling you the same thing in a multitude of different ways, but you refuse to get the point. Yes, the operation of the software we write can be affected by the physical realities of the hardware we use to execute them, but any time this causes the software to behave differently than it would in a purely hypothetical computer, this is considered an error and the results invalid. We even have hardware that automatically corrects for when rare events such as freakin' cosmic rays cause the value of a bit to flip (ECC memory).
The domain of software engineering is abstracted from physical reality. There's nothing useful to be gained from such discussions, because the whole job of hardware engineers is to enable us to operate at a higher level where we don't have to concern ourselves with the quantum electrodynamics necessary to make the transistors do their jobs.