HXX

Pretty sure there are starter kits that you can "docker compose up -d --build" that would be faster than AI, and if you've used them before you don't have to comb through the mountain of mediocre code to vet it (which you have to do every time AI generates a bunch of code for you), not to mention I doubt you clowns even go replace AI generated "secrets". The future is going to be a field day for hackers.

@rfleury @GrizzledTexan So does this mean you are conceding on the "two minutes of regular coding" thing or are is Larry full of shit?

Best anti-AI case I’ve possibly seen so far

@alkimiadev @JaidCodes It’s hard to imagine because you are a perennially mediocre programmer.

@JaidCodes Did the "clanker" or the human write that monster >4k loc file? It seems like the model probably did right by removing some lines in that dumpster fire file. Its hard for me to imagine a person who writes probably 5k+ loc files and gets butt hurt about a few removed comments

Thought this was an overly harsh response to someone who asks for permission to work for free. Then I looked into the PRs attached to the application. Bless Mitchell’s soul.

@ImperialHal @ZuniUK “If” isn’t that what you did?

@ZuniUK If it’s so OP just swap

The gap between controller and MNK has doubled with newer controller hardware being released, seriously considering swapping soon - MNK feels like beating a dead horse🫥

I mean, in 6s he appears as people mains in the top ranks at a 9.5x higher than random selection, and 5s tank role he is still represented 3.5x higher than 1:14 in the highest ranks, which is what you would expect if all the tanks were balanced. The reality is, he *can* be countered, but it has to be a conscious choice + people have to not ban some of his counters. My argument is that it's poor game design in the first place to be like "oh they have hero X, so sometimes (multiple people even) we have to swap so we don't get rolled". I mean my stance in general is much more severe than even that, I would remove Roadhog, Doomfist, Genji, Moira, Winston, Vendetta and every other high value, low skill hero in the game. I don't even think they qualify as "movement" heroes when the movement is trivial compared to any game with real movement (e.g., quake, tf2, apex legends), it's not a "skill" to press a movement button and then to have really easy to land skills and abilities once you press LShift or w/e. If those heroes must exist for accessibility reasons, they should be completely non-viable against any of the cast that has a hard aim requirement at a high level (one of the only mechanical skills you can express in this game). The inverted belief that heroes Winston has hard requirements on "positioning" are asinine, he has much more frequent movement and much easier escapes than other more stationary heroes. If you *don't* have movement & escape abilities you have to be hyper aware of where to stand and how to predict dives and use terrain ahead of time. On top of that, heroes like that usually have a hard aim requirement, sometimes even requiring landing a lot of headshots to secure kills in high rank games.

@NotAttained @DoctorMadrid10 @OWCavalry Then why is hog not meta? If it's SO strong and broken

New #Overwatch Balance Changes 🛠️ 🐽 Roadhog Chain Hook: → Cooldown reduced from 8 to 7 seconds. 👾 Sombra Stealth: → Movement speed bonus reduction while revealed by damage reduced from 50% to 33%. 🐺 Vendetta General: → Health increased from 175 to 200. (Total health increased from 250 to 275.) ⛓️ Mizuki Katashiro Return: → Enemies can no longer prevent Mizuki from recalling by standing on Katashiro Return's starting point. What do you think of the changes? 💭

Are you implying that having an easy to land giant hitbox ability on a 7 second cooldown that can kill most of the cast in 1 shot if they are within 20 meters, and even those it doesn't kill get displaced so badly they are almost guaranteed to die. Secondly, a lot of heroes have dmg fall off starting at 25 meters, you also walk backwards more slowly than hog can walk forward, so you are either doing less damage, or you are somehow taking a couple burst shots perfectly within that 21-24 meter range(lol good luck) then turning your back to enemies so you can stay out of hook range (lol). It's just awful design through and through, it's way too strong of an ability to exist, even on a 15 second cooldown, let alone 7. Vendetta and Sombra are also two of the most poorly designed heroes in the history of hero shooters, their kits (like hog's) are extremely easy to master and their abilities are easy to land, and they just result in a generally unfun experience for their opponents.

@OWCavalry I absolutely hate seeing people on both Twitter and Reddit whining about Hog just because they don't know how to play Ana.

I’ve been GM every season I’ve played, none of those heroes need buffs, in fact they are all horribly designed and unfun to fight against. It’s like they’re randomly in competition with Marvel Rivals to make their game as horrible as possible. If they were nerfed to a 0% pick rate it would be a great day for the game

@OWCavalry People mad about hog. He isn't even picked most of the time from the playerbase. His kit is outdated and rolls over quickly. We can use every buff we can get for both Hog and Vendetta. Sombras buff is good. Yall need to get better at the game instead of trying to banning her.

@gordonlyon I've been waiting years for the next one...

Tweet Tweet Tweet Tweet Tweet Tweet Tweet Tweet Tweet Tweet!

I think this coincides well with an issue with LLM generated code (on large complex projects), i.e., you don't feel the pain of the meandering, or poor API design choices, or poor performance choices early on when things are smaller & more simple. I would almost redefine the state space as being 3d with hills and slopes and that your state gets larger and carries more momentum, and is thus harder to redirect the line to go back up-hill towards the goal if you meander off a cliff somewhere. When there are 10,000-100,000+ LoC shipped weekly by LLMs, including tons of tests, it becomes infeasible for a human mind to keep track of, and to hand-modify things at any impactful level, especially when the state grows so quickly and has so much momentum. If it turns out that all of those directions are downhill, eventually there is a reckoning for how to get the state into the "goal area" when you have something massive that has continuously gone in the wrong direction down hill.

@cmuratori @teej_dv @trashh_dev youtu.be/Mi98zVBb6Wk?si… I found it!

LowLevel and Prime are gonna record something about password managers. If you have a question you want answered, reply here. Might be able to get it in the vid

Hey Casey, there is a video where you talk about the path a codebase takes over its lifetime as it is iterated on to reach some kind of goal, iirc there was a starting point and points where you showed samples being taken of the codebase being iterated on (moving in arbitrary directions), and how as someone gets better they also get better at not taking as many detours and moves in the wrong direction as they gain experience. I can't remember if that was a YouTube video or a substack video.

@teej_dv Everyone knows if you want password advice, you ask @trashh_dev

I would say that there is something similar to what you're talking about, and that it could be akin to adding another dimension of height and "slope" to the path your code takes, where if you really go off the rails, it literally becomes more of an uphill climb, of course it isn't linear like this or so black & white, but I think the gist of it is true, often times you can take so many steps in the wrong direction down the wrong "hill", and sometimes it's even easier to go downhill, but it becomes that much harder to shift the codebase back in a good direction especially when there is momentum behind the downhill trajectory. So you can meander down somewhat in the direction of the target you are looking for, but ultimately getting there becomes virtually impossible or at least extremely difficult depending on the scope of the project.

I can't find the video I'm remembering, but it was essentially like this You can move the state of your code in some direction each time you iterate on it, and end up taking a meandering path toward the target, the only thing I would amend to his "journey" analogy I will post as a follow up to this

Beyond parody

@yarnf40580 @rfleury Yeah, I think Casey Muratori had a good abstract visualization of a "codebase" moving through 2d space as a line, let me see if I can find it

There is a much more insidious part to all of this now too, the codebase moves faster than I can manually keep up with it, and the AI touches EVERYTHING, there are so many interdependencies and merge conflicts that it's not even feasible for me to keep my work if I don't manage to merge it within a few hours of starting, I often just have to nuke my branch and start over on top of the AI's work. It also generates so, so many tests... Which means it's a pain for me (as a human) to go through and figure out which ones are worthless and which ones aren't, but I *have* to do something, because it's an auto-reject on PR if tests are failing.

@NotAttained @rfleury You guys must already almost be done for the year, at those productivity levels

@yarnf40580 @rfleury Imagine now, working on projects with people who generate LoC in the 6 figures *weekly*. Life can be more sad than I ever imagined

@rfleury it was literally, for every single read, iterating through every single shard, acquring a lock every time, and cloning the entire shards entry-map every single time, despite only reading an entry thatd be in one of the shards. So ya, 4202 tests like that would be just great!

@rfleury @BrianSpragge "right now"... does that mean you're going to add that :)

@BrianSpragge It doesn’t have any editing functionality right now, it’s just a debugger.

Someone mentioned to me that they missed a file tree view in the RAD Debugger, so they could easily browse their project folder hierarchy. I then realized I had inadvertently already implemented it in the watch window.

@Acerola__t @DaisyCaladia Was it really a perpetual NDA & does that mean you wouldn't be able to implement any of those trade secrets into your own engine(s)?

Probably because Korean ESL pros who dominated native English speakers for so many years didn't understand what hitscan meant, and now everyone who got dominated by them also uses incorrect language... You guys got mogged by non-English speakers into forgetting your own language in Overwatch, thankfully other shooting games don't have this problem in the pro scene

@NotAttained @SkaVorah @Ja1de3nz We are talking about the role designations of heroes in organized play, not what they shoot. Her role is hitscan/main. This basically means you want to pair her with a flex dps. It just a label of roles in comps. It’s based off the type of pressure, not what they shoot.

I will just say 3 things. 1. Sierra absolutely MELTS Tanks now. 2. If she lands her hitscan-like dart on a you as a Support, you are pretty much done and deleted by auto-aim. 3. Ult nerf is not noticeable at all. #Overwatch

@foijanow @SkaVorah @Ja1de3nz And I explained that you’re wrong, if you mean e-sports team, no I’m not on one, if you mean do I play ranked, I’ve been GM every season I’ve played up to and including right now

@NotAttained @SkaVorah @Ja1de3nz I literally just explained how the bullet is irrelevant. It’s the TYPE of pressure they can output and what they do in comps. I’m assuming you’re fully casual? If so, these terms don’t really matter too much for you.

Do you think that running non serially dependent code across SMs/warps and doing programming as “wide” as possible will unlock huge speed ups for general software? And on a separate note, do you think that heavily graphical applications (like games) would suffer from from sharing resources for things that were typically more “traditional” compute, and on that note, if a lot of code is branchy/serially dependent, I’m sure there are *types* of computation that are generally faster on the CPU, including dealing with latency hit for VRAM access? Just random thoughts, Im sure i’ve seen you address some of it before e.g., explicitly making code less branchy

It’s still early in development but what we’re developing is a shader compiler guaranteed to maintain flat execution times on GPU regardless of the data inputs to ensure real-time deadlines are reliably met no matter the content or user behavior. You’ll be able to program it like a video game, visualizing a simulation of the virtual GPU executing in a manner similar to Minecraft redstone or Factorio conveyer belts. The idea is to enable making performance bottlenecks visually obvious. We’re also creating our own real-time debugging tools built atop a GPU-based interpreter, as well as a real-time multiuser development methodology that’ll act as an alternative to version control. Two markets we are targeting. One is general purpose software for desktop and server applications. The other more premium product is targeting the safety-critical embedded systems market. We still have some R&D to do but we are hoping to revolutionize that market and replace the need for custom safety-certified hardware, operating systems, drivers, and HMI tools. This might turn into a GPU-based RTOS, we shall see. The idea is to treat the GPU as a hard real-time coprocessor where the CPU is only responsible for being a middleman for I/O, nothing more.

Software engineering is largely solved btw.