I became a rockstar at work last week when I created an executable with claude that took a generated excel file and processes it in 5 seconds. This would normally take two people a little over an hour to do. My question is this, what happens when its all automated? Physical labor? You have people at work who all they do is send triggers to each other through emails, basically asking people to add routines, etc. When do companies start realizing they dont need people anymore? Ive heard only whispers, but have not seen it on a grand scale, but do they just want all their workforce just moving heavy objects? opening hundreds of boxes, etc? I see why all these companies are preparing for the worse.

why?

>>110102378
>>110102335

>>110102374
no one fucking knows and stop asking questions you're standing in the way of progress

>>110102382
oh...yeah RIP

>>110102378
4chan is a honeypot and even joking about developing potentially new, novelty weapon systems is the best way to end up with a black car in front of your house

>>110102383
Its progress if we automate it away, but what does an employee do with all that extra time. If the entire point is to automate my position away I honestly dont want to get involved, even if it is progress. Progress shouldn't mean I don't have a paycheck to eat food any more, and that's where this is headed. The clowns who think the government is going to come to the rescue are dreaming.

>>110102401
i never even thought about dual use the entire time i was coding so essentially it's clean. if anyone deep scans for anything that even looks like weaponized embodied robotics all they'll find is mathematical primitives and arxiv citations

File: 1767190648333604.webm (3.93 MB, 720x1280)

3.93 MB WEBM

>>110102374
The real question is what happens when human physical labour is redundant as well, which is following shortly after.

>>110102401
No it’s not at all, and your word salad has no potential to be a new weapon system.

>>110102422
nigger that's not even me you're replying to. "lmao". anyway, ask a question for the first time in your life. or are you too afraid of getting mogged.

>>110102431
Sneed or chuck?

File: skeletal.gif (955 KB, 360x360)

955 KB GIF

>>110102439
lil' chuqita

>>110102420
thats coming but i think thats going to be the hardest step, they would have to redesign the entire factory from scratch to accommodate those robots. just seems like a lot of labor is still needed, probably more now than before, only good thing working against that is labor costs, paying people to turn screws is expensive. I dont foresee that shit happening in at least 10 years.

>>110102420
imho, extreme riots that will make blm look like teletubbies. for their "right to slavery" normalcattle will chimp out

>>110102482
>they would have to redesign the entire factory from scratch to accommodate those robots
No they wouldn't. That is the entire premise.

>>110102482
>redesign a factory designed for humans
>for humanoid robots

The real redesign is when they go from human to robot arm, blud, this will be 10x easier.

>>110102431
Can you explain how your schizo slop does anything for drones?

>>110102528
on-the-fly autonomous discrimination between targets and noise in milliseconds without burning a single llm token in completely jammed ew frontlines. want more? this is just off the top of my head.

>>110102538
And how does it do that?

>>110102559
a ~10m lightweight param hrm/trm staged with general mathematical primitives that do certain things i don't understand because i'm not that type of guy. basically doesn't do matmul but just isolates what's going on, and what needs to happen. can solve arc-agi-3 puzzles and win entire games without calling the llm.

>>110102568
And what medication are you prescribed?

>>110102585
deez nuts.

>>110101864
The new knights. If you can't afford your own armed cavalry or the protection of a liege lord you're just gonna keep getting raided and raped.

>>110102507
>>110102506
Its not just the factory either, they will have to redesign the products as well to work with robots, so its more efficient and can be held easier, etc. Putting together things that dont want to fit together is whats holding us back and is why there are still people working in assembly plants. For example where I work, its not physically possible to switch people out with robots, as much as I would like to see that happening. Personally working for humans is slavery, some people will say "No ones holding a gun to your head blah blah", they might as well be, if you dont work in this country you get no money and then what? rely on the government who doesn't give a shit about anyone? Companies all share the same brain, and if word gets out that employees are nothing but check collectors, they will start getting rid of them, and then the next company follows doing the same thing.

>>110102383
this
whiney doomer should stfu

>>110102628
honestly most companies are too sluggish and are ngmi. you're not thinking big enough wagie, think about ubi and resource post-scarcity.

>>110102628
Robots can do the same fine motor control as us but even more precise and can lift more weight, they don’t distracted, they don’t get fatigued, they don’t get the shakes from substance withdrawal, they don’t have poor eyesight.

We will be made redundant soon.

>>110102639
that's where you are wrong, I want this to happen more than anyone, the problem is getting to that point, quick pain is the best pain, but if all we are going to get is more physical labor, and less skilled labor, what is the point then? just bring every pajeet and juan here. It'll be a world where all the idiots are rich and all the smart fags poor, which I have nothing against as long as I still have food going in my belly. Just saying, you guys are seeing me the wrong way.

>>110102671
Yes, maybe humans were not supposed to be smart, maybe we can focus on more important things, like bettering ones self, health, family, etc.

Free GLM and Deepseek from opencode running for almost 2 days now in search for a dreamcast vmu exploit and I don't think they gonna cook anything useful...

>>110102701
try using the new one from deepsneed api

>>110102710
But that costs money

File: aimscore.mp4 (545 KB, 1920x1080)

545 KB MP4

making progress on my game engine, now having a spergout with 5.5 about asset packaging, meawhile a game building as we work on it
an aim trainer mostly made to stress test the input system, we poll at basically hardware rate so if your mouse provides updates at 1000 Hz we get updates at 1000 Hz

and also working on the UI framework in the meantime

File: bf9ad3ce9c245c4ec8efda8df(...).jpg (547 KB, 1024x1010)

547 KB JPG

I talked to another student the other day and he was convinced that Anthropic and OpenAI need to start making a profit soon, and once that happens the vibes end and we're all going back to hand writing code.

I wish I didn't have to work and could spend all my time just reverse engineering everything and releasing it, people want a revolution, you can actually make it happen with this with enough people doing it and basically spreading the knowledge.

Does Claude Code have something like Pi's codemode?

i plan to sub for claude code for the first time soon, does it have a referral system i should take advantage of?

>>110102753
where is the sex

>>110102768
I think when you sub you get 4 ref links that give people 1 week for free. If they sub you get credits.

>>110102782
I'm too poor to setup local but that's the vision

File: 1761873276009574.png (105 KB, 562x838)

105 KB PNG

After using ChatGPT Codex exclusively for a while now, I finally decided to give Claude Code a shot. This desktop interface panel that has the context usage and breakdown is really cool.

>>110102761
https://en.wikipedia.org/wiki/The_Last_Question

>>110102721
curious whats the prompt? you should at least input some referense assembly a little bit "vuln" that they can handle.

Please respond lads:
I have questions about all these decomp projects that have been in the news.
I have an old windows game I have had my clanker chug through after it decompiled the exe and executables, but its been a pretty slow project that has had a lot of errors along the way (like when I test a function it claims was ported identically the game still crashes and we need to fix it)
Is this a case of bad prompting on my behalf or is it the fact Im actually trying to create an open source equivalent instead of just a wrapper to execute things increasing the complexity beyond what a lot of these high visibility projects choose to do

>>110102990
use linux.

>>110102990
model skill issue

>>110102990
It depends on which model you're using, they're not as equally capable. Secondly these days it blew up thanks to tools like REA MCP, since MCP is more token efficient and your model can do more with your sessions.

>>110103007
I have used various iterations of Claude and chatgpt over the last few months, I have not actually ever set up that rea project yet, but mostly it's been running things though ghidra and then having the AI work through the output.

>>110102420
If you're american, you're probably going to be OK. Guess out of where all the tokens to drive this shit are going to be coming from? Yep, our data centers. We just *might* get UBI considering how well armed americans are. Chinese will likely be fine as well, they will be using the robots and making them.

Other places are completely fucked. Europe will simply collapse, literal blood in the streets, they have nothing to stop this, its going to be an abbatoir in there. Famines, too, god knows what else. Maybe Russia will take the whole place over and remake it into a Soviet Union 2.0

so is it true that people are reverse engineering software just with pictures of the ui and stuff? how is that even possible.

>>110103028
>Europe will simply collapse, literal blood in the streets, they have nothing to stop this, its going to be an abbatoir in there. Famines, too, god knows what else
Show's what you know, the EU will just make collapse illegal, and will punish everyone that collapses with 20M or 5% of annual turnover rate whichever one is highest.

>>110103031
depends on what they have their agent doing. could give them a picture of a version of some software and the agent downloads it and works from there, could just copy what it knows that software does, etc

Does anyone have any recommendations for API resellers of flagship models below Anthropic's, OpenAI's, or Google's rates? I read about Chinese resellers who for example, pool Claude Pro accounts and then expose an API via that pool such that their offered API price is lower than what Anthropic charges, so I know they exist. Does anyone have recommendations for services that provide this discounted API reselling or know more information? Thanks in advance.

>>110103031
The AI assisted reverse engineering tools was already available last year. It was just waiting for a capable model

>>110102990
Number one question. Do you have a feedback loop for your agents?

using claude to make shit it blows my mind that it only costs a $20 sub. there's no way they're not bleeding money

>>110103055
A lot of their profit is from lying about what model is responding. Even Kimi does it, you think a chinese scammer wouldn't?

>>110102990
are you copy pasting with web chat?

>>110103073
Yeah we have a comparative harness that each phases works to match behaviorally. It's not that it does not work I'm just baffled that I have like a 3k function dll and I have seen the entirety of the ps5 seemingly get reverse engineered in the time I got like halfway though my project.>>110103080
It's all the standard codex or Claude desktop apps accessing the code base autonomously

>>110103096
Alright, my recommendation would be
>like when I test a function it claims was ported identically the game still crashes and we need to fix it)
this is where the inefficiency is.

you'll need to original toolchain (same compiler, the same version, and the same flags)

have your agents make a tool that compiles a function using that toolchain and returns a objdiff for the % matching, and they loop against that until they get 100% for a true match.

>>110103176
That's a very reasonable suggestion, thanks anon

What niche software exists that you would like to see reversed engineered. I'm not saying to do it, but what software once reversed engineered will it actually make things much better? I get the low hanging fruit like excel, but what softwares exist? Im sure there are crazy niche mining tools, like actual mining that I know exist which look pretty crazy but costs a fortune, same with maybe Catia or engineering software, but im sure there are other not well known stuff.

>>110103245
I think the biggest problems are where/how are you obtaining this software and disabling the licensing

>>110103253
Chinese sites have them if you search their engines, which I would not recommend, but they do exists. Ive never used them myself but I know people who have sworn up and down that a crack does not exist, and that it cant be cracked because of a usb key, but ive seen it and its real. Last one I saw, which I have not tried involves using everything and replacing a file it generates, then going back to the installation and installing the software, which then they replace exe and its cracked. Im sure military software exists but would not touch it with a 100 foot pole.

>>110103253
Rutracker

>>110103245
things like what farmers use, waterjet cutter tools, theres a whole world of niche shit out there i have no idea how you would find it though

>>110103245
Holley ECU operating system and tuning software

>>110103245
IDA Pro Ultimate

>>110103298
I know of one software where you can reverse engineer parts just by scanning it, one might think that sounds really easy, but its never perfect, it involves a ton of work. So when a company came out with a software that basically fixes all of that, some corporation came and bought it. Im sure there are other priority software that exists which will be reversed engineered, not by me, but someone will do it. Its like 3d printers no? Didn't they start off with some proprietary technology that basically put 3d printers on ice till its patent ran out, imagine that happening but on a larger scale, who knows what these people are hiding behind closed doors. This is why the left are stupid and dont see the potential, a true revolution, which they unironically dont want.

>>110103245
Minitab

File: image_2026-10-18_131627397.png (502 KB, 800x600)

502 KB PNG

>>110102334
anyone knows a way of using the AI only to assist you as a reviewer and info searcher thing instead of letting it do it all for you? I really want to learn how to make a game engine myself, or at least understand how does it work.

>>110103360
self control

>>110103360
Let it go anon

>>110103360
use web chat, i.e, in your browser, instead of a harness like codex or claude code. i think all of them have live github integration now and accept most code filetypes.

>>110103360
yes anon, just ask the llm to do exactly that, an ai won't start coding unless you tell it to
and you can even write what you want from it in a file called agents.md and they read it automatically when you start them up so they always know what you want from them

Ok but when Fable 5.5

>>110103375
this is terrible advice btw
web chats have fast cache eviction in general and they don't get access to local code, so how can they explain how something works if they can't see the thing and trace the internals?

File: file.png (185 KB, 929x477)

185 KB PNG

>>110103388
this is my local code on the doorstep of a full linux vm and multi turn agent, not even close to as good as a harness but it will do shit

>>110103405
oh, that repo
figures

>>110103423
i would never let web argon touch my code after trying it once. never again anon. research partner and reviewer only, just like the frogshitter wants.

>>110103433
>i would never let web argon touch my code after trying it once. never again anon. research partner and reviewer only, just like the frogshitter wants.
lol this has been my goto for years with gemini now, everytime I tried it it always messes up my stuff and I need Codex and Claude to fix. Im never again trusting that POS now matter what the benchmarks say it might be AGI/ASI galactic level IQ or something never again.

>>110103456
argon is, literally, the first gemini model that's actually worth something technically & functionally. and i am saying this as a shill

>use Haiku they said

>>110103405
whats a pirates favorite LLM

Where does "Argon" stand between Pro and Flash? Did Google forget what model names mean?

>>110103505
jev

File: Untitled.png (79 KB, 1311x722)

79 KB PNG

Did this a while ago because I couldn't find any Conways game of life + music tracker combos that did what I wanted. Pretty simple, but has some neat features. Can change the GoL rules, and limit the playback to certain scales and such. 4 instruments. on tabs.
It's not a music production app or anything, but you can save the music as a json file. Comes with an example song you can load frog_head.json

https://files.catbox.moe/hz1bu2.zip

>>110103509
stands head and shoulders above the old pro line, completely different architecture. i understand why they deprecated

>>110102401
we have a literal weapons board.

>>110103539
run by nafo

im abit confused, i added "add text context" to the Context but theres no way to edit the .md after creating it, is that intentional?

File: 1.png (2.06 MB, 1536x1024)

2.06 MB PNG

the silence is deafening. can we all agree tech stocks are going to get raped to non existence? I get the AI chips parts, but if you are a software company you must be shitting bricks right now.

>>110103588
we dont know what you're working in

>>110103597
oh sorry, claude

is there a reason to still use claude.md nowadays?

>>110103600
-_-

>>110103611
yep the harness still has preference for it despite what anyone says

>>110102334
>## Best free tier
>https://opencode.ai/zen
I'm currently using the Zed editor with google's free tier gemini 3.5 flash lite. Is this zen the same thing as gemini where I can just put in the api key? Also how does it compare to flash lite?

>>110102950
I just gave him the bios image and said to find an exploit that works by abusing bugs related to data on the vmu, not by fucking with the maple protocol. It disassembled the bios with objdump, found potential routines that read icons, custom backgrounds or other card structures and it's now testing different memcard images in flycast as a debugger.

File: Screenshot 2026-10-18 130158.png (56 KB, 760x428)

56 KB PNG

thats it for me

>>110103245
Yeah all the mining stuff, iogas, leapfrog, micromine is super expensive, I’m actually making a gis/logging/admin all in one modified QField for exploration because I I’ve dealt with shitty excel spreadsheets, online folders and actual pieces of paper and I know exactly how it needs to be built.

I’ll probably do reflex/core plan next, everything tech related is paywalled and expensive as fuck, gyroscopic surveys, azi tools, plods and none of it is particularly difficult to implement, it’s just everyone in mining is actually a mouth breathing retards, overconfident finance bros or decrepit boomer psychopaths.


Starlink becoming ubiquitous has also been perfectly timed for the sloppening in the bush.

File: 1792338621499049.gif (3.05 MB, 342x498)

3.05 MB GIF

>Argon make agy useful, make no mistakes.

>>110103664
zen gives you quantized smaller models but they will still perform better than fleshlight bro.

>>110103713
And it looks like it says it found some potential vulnerable functions with OOB writes but nothing that would lead to code execution for now. It failed on some approaches like the driectory structure, FAT, etc. I'm not sure how much of this is hallucinated

What do you do when your free OpenCode usage runs out?

>>110102374
>My question is this, what happens when its all automated?
If we're lucky, a lot of death.
If we're unlucky and it happens after humanoid robots, a *lot* of death (of you and all your poorfag friends)
>Physical labor?
lol

>>110103905
>>110103932
I just restart my router when it says I got rate limited, I guess I'm getting banned soon

File: file.png (9 KB, 684x114)

9 KB PNG

how do i poke it to see whether anything is happening
context doesnt seem to be changing either

>>110103505
Arrrgon

>>110103960
i would terminate. my xp is with claude and argon and they don't do that unless it's really fucking serious business

>>110102701
>dreamcast vmu exploit
Just in hopes of finding a new one or what? Is there a more specific goal?

File: 1789027381117752.png (2.14 MB, 1248x1824)

2.14 MB PNG

another day of vibelording. haiku still goated. dipsy still goated. sonnet still goated. opus still goated. dario still goated. sam still coaled. astra still coaled.

>>110103997
I have a VA2 model where burned disc exploit is patched and the only known way to bring this back is to flash the bios which I want to avoid

>>110103869
>I'm not sure how much of this is hallucinated
that's what doing real testing is for, clears out hallucinations when it realizes on its own they don't work

>>110104017
Oh I see - good luck, anon, noble goal.

Do any of you even use this shit like grok bots and chatgpt dots? What’s the fucking use case? They’ve been shilling these products for weeks now on X.

>>110104113
yeah no clue I also keep seeing the shilling, but can't understand what it's for.

>>110104113
I use ChatGPT dots to give me weekly reports on the whole decomp/recomp scene. That's about it.

>>110104176
chatgpt.com/scheduled can already do that

>>110104113
No clue, I can already do everything those things offer in other ways.

>ask opussy (high) for help with brainstorming, the input is a dozen bullet points with ideas
>it writes a 90kb design document
jfc never go above medium

>>110104343
how much of it is tests/gates

File: 1789135081801769.png (244 KB, 1629x1029)

244 KB PNG

>opussy using rea to shore up xdm's weaknesses
>dipsy documenting everything for future improvement
oh yeah. it's all coming together.

>>110104410
we have to go deeper

>>110103031
They're lying for legal reasons if it's a complex app.

>please simplify this section
>simplify it further
>explain it like I'm a retarded 12 year old

>>110104565
unironically this, why the fuck do i need a low level explanation of how method X interacts with subsystem Y and how load bearing it is, just tell me i/o and a brief overview of how it processes it

how can I see how much money I still have on opencode go for each model

>>110103960
Just ask. The reason why they aren't showing you any summaries is because you might be Chinese.

astra LOVES missing the point and doing the exact opposite of what is good maintainable software design.
is claude any better in this regard?
my hate for anthropic has prevented me from downloading cc as of yet.

>>110104638
>hate Anthropic
>love OpenAI
>both are Jewish
Make it make sense

File: file.png (73 KB, 1097x825)

73 KB PNG

>>110104638
i find sargoon to be about as good and readable/maintainable as opus 5.5 if not a little easier and less pedantic

>>110104638
unfortunately it's much better.

>>110104638
for example?
depends on what you call maintainable
I'm working on some projects, I can show you the code/architecture if you want, it's all opu55y

Kimi Code nowadays is fucking dogshit... I only got 170k tokens before hitting the 5h limit, less than an hour. I'm never again subscribing to this shit or getting excited for new chink model, fuck them.

>>110104685
kimi k3 was always C tier

>>110104689
Wrong, it was better than Sol on release day.

>>110104694
>chinese thing looks good out of the box
many such cases

>he fell for the chink model meme
>bu-bu-but the benchmarks
how many times do you guys need to be told that chink models are shitty distillations of claude with benchmark RL and extremely shitty real world use performance

>>110104664
I find your cole highlighting to be crime against humanity

File: 1465008348731.jpg (33 KB, 453x500)

33 KB JPG

>>110104734
it's notepad++

File: 4uwisi3cr9wh1.jpg (157 KB, 1122x1308)

157 KB JPG

which one of the companies seen in this picture will be dead in 5 to 10 years?

>>110104732
kimi and deepseek have good real world performance, although kimi is old and too expensive to be getting mogged. it needs an update

>>110104760
adobe

>>110104760
Its about time, some of them have had a monopoly for decades, its time to go.

Opencode is also rugpulling a bit by banning other harnesses and adding higher priced subs.

>>110104732
Kill yourself dude, never looked at a benchmark, K3 is/was actually good, it's just that the new limits are unusable.

>>110104706
OpenAI did the same thing, their $100 plan is nearly unusable.

>>110104760
sap and salesforce most likely given how greedy those companies are for what they provide. matter of companies enrolling their own CRMs when they realize how much costs they can cut

>>110104763
I used kimi once when it was brand new. It was barely usable at all. It took forever and there were no good plans for it, that plus being a token hog it ended up being retardedly expensive, slow and stupid

>>110104760
Anthropic will probably be dead, same as cuckAI. AI as technology will be alive and well. Hardware will still be expensive.
Some SaaS survives, others die. Adobe will live but be much poorer. Enterprise trash like SAP will inexplicably be fine, to the bafflement of all (including people working at SAP). Most consumer facing SaaS will be dead.

everything Chinese is garbage

>>110104802
don't talk shit about my wife

>As far as I remember (I’m not 100% sure of the finer details):
stupid fucking clanker you're a ROBOT you don't remember anything GO CHECK instead of "lole i forgor"

File: 165637B2-F8B5-4963-B895-7(...).jpg (157 KB, 3144x1584)

157 KB JPG

it’s fucking OVER

>>110104798
Nah they're too big to fail, the US government will bail them out.

>>110103753
What software do gynecologists use i have an idea

>>110104825
This is obvious to anyone who uses both.

>>110104798
Adobe is bigger than what they provide to be honest, these companies have had monopolies since the 90s and 80s even, and nothign there to stop them until now. They will be long gone before the 2040's, or shells of their former self. SAP will likely still exist because companies have been using it forever and are stuck with it, but who knows, they might pull the plug on all software and just design their own. SAP is bullshit anyways, using stupid transaction codes when everything should be right there in front of you, no more complex software to lock people into ecosystems.

>>110104827
Anthropic will be the name of a country in 10 years

>>110104825
how the fuck did they do it bros? the 5.5 models are insane.

>>110104825
I really don't like claude and I'm afraid that when I switch gpt will beat it

>>110104850
https://en.wikipedia.org/wiki/Project_Panama

I don't like claude because it has a personality that it inserts into everything

dont be surprised if you start seeing all these tech companies drop back into the early 2010's as far as stock prices go.

>>110104861
buying and scanning used books destructively oh no

>>110104878
yeah one of each copy, not that bad. i ended up talking to sargoon about my grandpa and it quoted his book from memory, the chat led to us figuring out from the online book provenance that google has deals with universities to scan non-destructively.

that reset was useless
my tokens got chugged in a nanosecond

>>110104780
>Opencode is also rugpulling a bit by banning other harnesses
I know the free tier only works in OpenCode, but I haven't heard anything about them restricting the paid plans.

https://opencode.ai/go
>Go costs $10/month. You can use it with OpenCode or any agent.

>>110105007
I really don't know what opencode is.

Does it do the new deepseek?

>>110104932
I've been running hermes nonstop on an endless goal and you finished your tokens already?
I'm only at 22% left of weekly.

I will always find it hilarious that Anthropic and OpenAI complain about the Chinese stealing from them while themselves steal shit

>>110105059
>hermes
???

I thought that was a harness.

Argon has to be Google's worst launch yet. Yes, the model is a good model, but it's crazy.

I just had Argon on max effort on the web thingie have an unspecified no number error mid-reply.

>>110102334
Cyber Chase on PC when?

>>110105089
the only thing that matters, the only real datapoint, in your entire shitpost, is that you got an error. cancel your sub.

>>110105089
>Argon has to be Google's worst launch yet.
most people just masturbate to bench[od]maxxed colorful rectangles and then go back to using claude or chatgpt
saargon had big rectangles therefore it's good

>>110102334
I wanted this videotape so bad as a kid, I got in trouble in school for just writing "video" over and over for an entry in the journal they had us write in every morning

I got it and I watched it a lot, it was good movie.
It skipped a lot of stuff though

>>110105089
I'm pretty sure it's the only model that can actually listen to audio though which I have found incredibly useful.

>>110105097
you can do it yourself

File: gigazatt_reel.webm (3.11 MB, 960x540)

3.11 MB WEBM

What should I add to the game Anons?

>>110105050
>I really don't know what opencode is.
The OpenCode name is used for both OpenCode (the harness) and OpenCode Go / Go Plus (the model subscription plans offered by Anomaly, the company that makes the OpenCode harness).

>Does it do the new deepseek?
DeepSeek V4.1 Flash is included for free (with daily limits) in the OpenCode harness, and it's also included (with $80/$160 allowances) in the Go / Go Plus subscription plans, which can be connected to any harness. I don't believe they're serving the 1017 update yet, if that's what you're asking.

File: 1770720767854992.png (96 KB, 738x349)

96 KB PNG

>>110105183
its over

>>110105100
>datapoint

ow, got me right in the aids.

File: gigazatt_reel.webm (3.12 MB, 960x540)

3.12 MB WEBM

>>110105201
gah +1000 Izatt out of the gate.

File: IMG_8875.jpg (22 KB, 717x579)

22 KB JPG

I’m not too upset about AI killing programming. I’ll be more upset when AI interfaces to the kernel improve and kill shell/terminal environment.

File: ORB.jpg (45 KB, 680x680)

45 KB JPG

>struck with an idea
>an Elder-Scrolls HoI4 mod
>have a good idea of what systems to look at and modify, and how a pre-production meeting would look with The Council
>half-way through a different project porting mario into san andreas

>>110105267
this would only maybe make sense if you're a windows or mac user. which you shouldn't be anyway if you cared about that at all. even then I don't see winblows or mac getting rid of the terminal.

>>110105267
That's stupid if there was something better than CLI we'd use it too, what do you want a bash MCP?

I need to get a case of those rtx 5090's.

>>110105299
at least elder scrolls hoi sounds interesting even if uninspired
mario in san andreas is just retarded and would only work to farm engagement in reddit and x

File: 1792355851.png (129 KB, 2526x468)

129 KB PNG

tfw opus is cheaper than deepseek

I think all agents should have a persistent python/node repl tool. Or maybe code mode should also work as a repl.

>>110105634
pretty sure marion in san andreas has also already existed for a while

>>110105174
wait it can? like actually analyze music tracks and answer questions about it?

i'm gonna blow your little minds!
imagine...
mario in sonic
yeah...
*drops mic*

>>110105637
>Unable to use the tool; limited to novel writing/tavern users only
ouch ouch ouchie ouchie

>>110105267
AI is literally based on WORDS
it won't kill the terminal which is the fastest way to process words

Is there a truly comprehensive, autonomous skill.md that can continuously analyze, understand, improve, debug, refactor, and repair an entire codebase without requiring me to manually identify issues, write task lists, or specify what needs fixing?

>>110105634
It was more a proof-of-concept or perhaps proof-of-ability when using claude, as the tools to do this already exist (libsm64) and I have no doubt someone else has already done this to mystify normies.

I just wanted to see if a team of agents could do it, with little-to-no human input beyond testing/logs.

>>110105174
yeah, I've been using it to properly integrate existing music into videos. Opus can only look at the spectrogram which isn't good enough.

>>110105703
>>110105660

>>110105179
Can't someone else do it?

>>110105694
kek
LLM are based on words.
AI is based on whatever you feed it (image for image GAN, sounds for music AI, etc.).

>>110105305
>I don't see winblows or mac getting rid of the terminal.
Not even two years ago everyone was saying manual programming is safe, now the writing is on the wall and it’s a matter of adoption. Prose text (and subsequently voice to text) commands as a primary input will only get better and will inevitably supplant the explicit typed shell commands we use today.
>>110105394
MCP won’t even be necessary, the functionality will be absorbed by models with better post-training

>>110105244
Sovl
Can you add more NPCs?

>>110105737
Tf are you talking about? Multimodal output tokens for kernel instructions?

>>110105737
the terminal and shell will always be an option. don't do false equivalencies because this has nothing to do whether or not snailcat programmers will get replaced. if you actually gave a shit you would just use linux or bsd but instead you're pontificating on crap coming out of your ass.

>>110105773
NTA and I'm sure it will always be an option but I think it's silly to think that our current stack will be the predominant one once the terminal layer becomes the main bottleneck for agents. More than likely there will eventually be a completely alien OS stack that is not designed for human use.

File: Screenshot 2026-10-18 170336.png (611 KB, 964x564)

611 KB PNG

My Fate/Extra CCC Vibe-coding project called "Project Kisaragi-SWORD" has hit the title screen!

only took a day

someone should put REA to work on finding out how to actually use CVE-2026-82079

>>110105696
Would San Andreas benefit from a decomp? I know that the modding scene is huge for it. There's a brazillian guy making mods to modernize the engine to this day. Proper shaders I think

If i wanted to slop a mod for a game, do i NEED Claude/GPT subscription or can i get away with using some free options like DS or kimi? I tried doing some stuff with deepseek 2 years ago and it was painful...

>>110105912
https://github.com/gta-reversed/gta-reversed
it's also had the real source leaked

>>110105929
deepseek has come a long way. it's something you'll just have to test. i mod slopped a mod for halo 3 with opus 5.5 medium

File: VVV.mp4 (3.69 MB, 720x892)

3.69 MB MP4

GOD CODING

What’s the best SI for game dev?

>>110106085
>What’s the best SI for
Opus

>>110106085
GROK

>>110103077
>Thanks
Dang okay, no free lunch I guess.

>>110106085
Sol because it actually does what you say instead of over-engineering

>>110106038
neat

I have a bad feeling Opus-5.5 is the peak of consumer AI capabilities. There will be a few more powerful models kept private in the name of safety, but I'm thinking we've reached the plateau. Can't shake the feeling.

>>110106176
I just need 3D to get a little better. Coding is whatever.

>>110106176
how would you even beat opus anyway? it can do anything, maybe in speed only
i think what will happen is that the jaggedness will be smoothed, instead of improving at coding models may get better at writing prose or doing law or physics or other stuff
coding is, i would say, something like 95% saturated or so by 5.5

>>110106176
Can you FEEL the plateau?

>>110106176
The Claude 5 hour limit is the pinnacle of human achievement

I hope you're enjoying your black fruit. The day of reckoning is coming.

>>110106176
Look at how much a gigabit of internet usage would have cost you 20 years ago. I'm inclined to think things will only get cheaper. Safety maxing laws are the only real risk.
Though as >>110106185 says I'd be fine with this being where code ability peaks. Just make it faster and keep making it better at 3D modeling. Not that I will complain if coding gets better too, I'm sure I can keep scaling up my ambitions to match

>>110106185
It could become reliable enough to automate office work. If it stops now, it basically just cucks artists and computer nerds (the two most mentally fragile populations) and leaves everyone else alone.

>>110106176
I was absolutely being routed to a newer model a few days ago. The thinking traces were 1/5th the size of it's normal thinking, and it was a step level increase in capability. Maybe Opus 5.6 or Fable 5.5.

When I got routed back to Opus 5.5 I could feel the retardation creeping back in.

>tokenmaxxed on a project
>ran into daily limit
>bought codex to help me finish the 5% left
>daily limit reset
>sudden realization that i don't hit daily limit on most days
>now i have to hit daily on codex
fuck

>I will discover the Task tool schema and launch agents with precise scopes. The adjoint mismatch likely stems from make_input during get_graph, noncontiguous tensor multiplication, or parameter aliasing. I will remove the debug logging once the root cause is found.

This is a great example of the snailcat problem. That's a whole meeting right there lol

how do i reduce token use on claude

>>110106233
>it was a step level increase in capability
How so?
People keep saying Fable 5.5 is liable to drop any day now

>>110106252
/plan
set clear goals
detailed spec
/compact
https://github.com/dmtrKovalenko/fff

>>110106271
thanks boss

>>110102374
The real thing is you don't tell anybody that you're automating your work and pretend like you're still doing it by hand, thus getting paid to sit around and finger your asshole.

>Found the cause:
...
>[x] wasn't the issue.
It's been doing this for 2 and a half hours now.

>>110106176
You would have said this about Opus 5. And 4.8. And 4.6.

>>110106253
I'm working on a project with 3 months worth of meticulous documentation, thousands of experiments with tens of thousands of notes and logs across thousands of files and folders. Whatever model I was routed to was seamlessly able to correlate across all of that, referencing small experiments in July, August etc. and compare them to recent ones to draw novel insights. Every other model I've used (including Opus 5.5) is simply not able to efficiently retrieve information like that. I have to point it to certain families of experiments and explicitly ask it to try to draw correlations. And usually, even when I do that, it gets context poisoned and quickly forgets the current experiment's relation to old ones etc.

Burned through an entire week's Codex Pro $100 in 16 hours. Most of it was Sol too. Friday my usage went from 40% to 0% in 10 seconds. I only keep an OpenAI account for imagegen and to have another model from Claude. Resets every 3 days dont even make up for it. OpenAI is a joke now.

File: Opussygame4.mp4 (3.92 MB, 800x430)

3.92 MB MP4

>>110106085
>SI

I'm loving Opus 5.5 so far. Managed to add a stalker enemy to my game with a single very long prompt, and a bit of correction afterwards. First time I feel like shit just works with LLMs.

>>110106344
You think it peaked now but you didn't think it peaked before? Curious.

File: file.png (101 KB, 634x440)

101 KB PNG

codex continues to not give a fuck, lmao

test

>>110106365
Which game engine is that?

>>110106410
Put Mario Odyssey on Firefox

>>110106410
It just has really weak guardrails it seems. I'm generating so much porn with it using my comfy harness

>>110106410
I might consider buying a switch 2 if I can jailbreak that shit like the 3ds.

File: Opussygame5.mp4 (2.39 MB, 800x430)

2.39 MB MP4

>>110106414
Bevy 0.20

>>110104848
>Anthropic will be the name of a country in 10 years
Or a moon colony

>>110106342
I kind of like when that happens because it feels like I'm actually doing something when I guide it towards a solution. It's especially satisfying when some vague intuition about shit I barely know anything about turns out to be right. I don't have to do the grunt work to figure out the details but I still get the satisfaction of solving a complicated problem.

>>110106410
is that 5.6?

>>110106493
sol 6.1

>>110105773
Why the strawman? I didn’t say shell won’t be an option. Manual programming will always be an option too, doesn’t mean it isn’t going to be an unnecessary encumbrance for 99% of users. See: assembly

>>110106555
It's soccer you uncultured swine.

File: 1742438882286737.png (631 KB, 1280x833)

631 KB PNG

>WATCH OUT!!! YOU'RE APPROACHING YOUR WEEKLY LIMIT! GET MORE CREDITS NOWWWW
>Check
>26% used
What a crybaby.

>>110106425
>Modern GPT-written porn
Yuck. Not even worth it. You're just gonna get

>You love it.
>I do NOT love it.
>--You do.
>--Okay, I KIND of like it.

>>110106608
I'm talking about images