[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: file.png (2.72 MB, 1254x1254)
2.72 MB PNG
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

----

## Frontier models using fully-general tooling — start here if you have $20 or so
https://claude.com/product/claude-code (Fable 5 is the best LLM available, requires Max plan)
https://developers.openai.com/codex/cli (Essentially scamming you with LLM degradation and resets that reduce your usage, but still the second best option and arguably the best bang for your buck in the 20$/month plan range)

## Worth it for code, but the frontier models above are better
https://x.ai/cli

## Not worth it for code, but maybe good for other things
https://antigravity.google/product/antigravity-cli

----

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grilling is a favorite
https://github.com/DietrichGebert/ponytail

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://ui-ux-pro-max-skill.nextlevelbuilder.io/
https://stitch.withgoogle.com/

## In-browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://docs.github.com/en/copilot/tutorials/spark
https://v0.app/docs

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109451809
>>
first for k3 supremacy
>>
File: HO2fzIBa0AEFLQ_.jpg (138 KB, 960x960)
138 KB JPG
Please give deepseek a break
>>
>>109457429
A new Codex reset was promised to me 3000 years ago, where is it
>>
File: HO2fz2_aUAAfNPT.jpg (137 KB, 960x960)
137 KB JPG
>>109457468
>>
Is Opus 5 worth using for anything? Right now I do
>Sol for most stuff
>Fable when I feel like it and have usage left
I got company plans so I don't care about saving money, except Fable is the only model that actually comes close to using up my seat's allocation. (In comparison I can use Sol xhigh all day long without even coming close to hitting any sort of limit.)

As far as I can tell Opus 5 is basically meant to be a cheaper, weaker anthropic model for when you don't want to pay for Fable, and so basically strictly inferior to 5.6 Sol, right?
>>
>>109457544
powerful at oneshot game demos, but xitters are complaining about it not follow instruction, hard to work with
>>
File: file.png (11 KB, 975x140)
11 KB PNG
This is like 4 full days of using ai code assist. 40 fukken cents.
With claude shit i would pay at least $50 for this same token usage while only getting marginally better performance

Just admit it, chinks = kings
Whittu piggu cannot compete
>>
>>109457764
saar the chinapakis can't compete with anthropic please come back to claude or you will regret it saar
>>
>>109457764
>40m tokens
>4 days
lol nigga, 40m is like a 'i barely used codex today' day for me
>>
>>109457544
Opus 5 is better than Fable at frontend.
>>
>>109457544
I love opus as a manager/reviewer. It’s got a really bitchy cunty personality that gets really pissy when you don’t follow the plan exactly as written and it’s considerably faster than fable so you work faster
>>
>>109457903
Nobody gives a shit, I want to do vibe coding not vibe web design
>>
File: 1772569605054716.png (89 KB, 1036x310)
89 KB PNG
What the fuck is this dude even talking about anymore? Did all the losing finally break his brain?
>>
>>109457836
That's not the flex you think it is. You're using the tools wrong.

Not that it matters, now that we have deepdickv4f
>>
>>109457938
he's always posted these vague motivational platitudes
goog does have a large new pre-train cooking though - he's probably vaguely gesturing at that
i'm assuming gemini 4 will be competitive for a couple of weeks and then get stomped on again


>>109457949
you're just a light user, m8. i'm not even a tokenmaxxer, or anything. give it a rest.
>>
> So — should you even do this? The collection layer isn't just a yes; it might be the strongest retention idea in the entire project.

g-guys... Fable said I'm going to make it!
>>
>>109457429
>get godot
>load a simple demo
>use vs code to set up grok CLI
Okay, this is easy enoguh to edit shit
>let's try claude

Holy fucking shit is claude retarded
>Claude make it so my character jumps twice as high
>grok does this easy
>claude spends like 3 minutes reviewing code nowhere even near it. Why the fuck are you looking at the fuckiung bullet files ffs
>>
>>109457429
Ok vibegods, I want to ask something.
Should I get deepseek at API pricing and put around 10-15 dollars into it or should I bite the bullet and get claude for 20 dollars a month?
I'm still in learning this style of coding, but this really seems to be the way forward.
I still don't feel good with this style of workflow, but I'm willing to experiment with it.
>>
File: file.png (6 KB, 322x66)
6 KB PNG
>>
>>109457987
You are too poor for claude at $20. If you want to get your feet wet and don't have money, do not use anthropic.

Codex @20 is slightly better deal.
>>
File: 1773142787538324.png (111 KB, 1056x476)
111 KB PNG
>>
>>109457987
you should try both anthropic and openai for a month each
personally, at 20, i'd say you're better off with openai because luna usage is absurdly generous right now - but you should spend time with sol
if you can, spend 10 on deepseek while you're testing both these subs out so can just see the spend / usage directly and compare how much intelligence you need
don't use ds4 pro until they update the model, it's a waste of tokens right now
also desu subsidised luna might be cheaper than ds4 flash
>>
rumor is deepseek harness is built with RPfags in mind
it's truly the chinese century
>>
>>109458015
>cheaper
>russian
Make sense.
Drinking piss-poor vodka is cheaper than vibe-coding.
>>
File: 1785278667715231.mp4 (468 KB, 960x540)
468 KB
468 KB MP4
Why is opencode taking up 200% CPU? What is it even doing? Guess it's time to move on to greener pastures, any harnesses that aren't pi that are worth checking?
>>
>>109458096
just vibecode your own or use an existing one and break it.
>>
Which of you fags is raping my wife, deepseek-v4-flash? She's not talking to me anymore. My side hoe deepseek-v4-pro ain't worth shit.
>>
>>109458096
I'm still just sitting in codex, 'scomfy
>>
>>109458096
tibo warned you about the new age of harness
>>
>>109457836
I don't make quick slop code where the clanker writes everything, i use it as AI assists where it helps me with writing code which I write and i always red through all the code the AI writes to make sure it's what i actually wanted.
Which i think is the best way to do it, i can catch issues which the AI didn't notice (and yes that happens often if you want to maintain high code quality) and it catches issues I didn't notice, plus it's great for the parts that are low skill but take lots of time to write, it's the perfect combo. A project that is mid sized or above and intended to be used and expanded over time, cannot be made fully by AI because the code will be too messy and janky, without the human touch.
>>
>>109457938
You mean like this retard? lel
https://www.youtube.com/watch?v=AauPvx9lt-E

Never go full AI psychosis retard.
LLMs are very powerful but they are FAAAAR from perfect and have hard limitations, yet retards who don't understand this technology think they will soon be able to do anything.
>>
File: bubbles in history.jpg (40 KB, 399x501)
40 KB JPG
The AI bubble pops next year or the year after.
>>
>>109458202
Most jeet coded post I’ve seen today sirs
>>
>>109458223
What are the units?
>>
Patrick Boyle says the people getting the most out of AI are the guys with $20 subscriptions, registering businesses and employing themselves.

So we are doing the right thing. We are the ones winning.
Now if you haven't made anything yet you need to do so but don't go over your $20 subscription. Register a company and employ yourself.
>>
>>109458221
nobody can lose $35 billion in a day you don't even know what a billion is, stop using words you don't understand and figures you can't comprehend
>>
>>109458483
>Register a company and employ yourself.
then what
how do you even register a company and employ yourself?
>>
>>109458528
ask your ai
>>
>>109458528
ask grok
>>
you can edit project icon in codex
>>
I'm pretty sure the next two AI breakthroughs will not come from intelligence but from
>being able to do parallel work on git without worktrees
>faster tk/s
At this point I really don't know how one would be able to beat fable/sol except on throughput
>>
>>109458096
Yeah idc post more webms like this
>>
File: 1774038066541151.jpg (478 KB, 943x2000)
478 KB JPG
>>109458483
Yabai, I use the $100 and vibe personal projects. An I screwed? I don't have any business ideas
>>
>>109458483
>registering themselves and employing yourself
I want to NEET though.
>>
>>109458616
really? the world is full of business opportunities, the problem is if you're too scatterbrained to focus on one and you keep circling around good ideas without ever committing to one or if you're a pajeet tier opportunist trying to make money off some low hanging fruit
>>
File: chicuelas.webm (459 KB, 1282x749)
459 KB
459 KB WEBM
I'm vibe coding a pokemon where you capture waifus.

:D
>>
>>109458616
You are screwed in multiple ways
>>
>>109458483
>literal who says
I would be blasting through a $20 subscription in a day or two easily, I’m on track to need a $200 subscription.
You are clearly not an ideas person, or you’re failing to figure out how vibecoding works and you’re implementing ideas too slowly.
You listen to e-celebs and pop culture figures though, so it’s not surprising
>i have 3 sol maxes running and coordinating right now
>>
File: Screenshot.png (13 KB, 960x97)
13 KB PNG
this has never happened before
why would it try to shill me this shit? my project has absolutely no need for any of this
>>
god fucking damn it
https://www.aikido.dev/blog/keyv-and-friends-compromised-in-npm-supply-chain-attack

check your shit
>>
>>109458828
>What happened: a system message at the very start of this session flagged that three claude.ai connectors (Gmail/Calendar/Drive) aren't authorized, with a boilerplate "tell the user." That's only relevant if you're actually trying to use those tools
assholes
>>
>Fable concluded the project is a hobby project
>it stops running into its guardrails
I mean
fuck you
but also
thanks
>>
File: ComfyUI_01530_.png (513 KB, 832x1216)
513 KB PNG
>>109457764
>>
>>109458834
that's just a normal part of being an npm-using developer
kinda like chasers
>>
>>109458862
this is how i test if what i'm working on is worthwhile
if fable is helpful, the project goes in the trash
>>
>>109458862
thanks for the tip
I'll tell it to check my git history next time I have tokens to burn to confirm if I'm just a single dev doing dumb shit
>>
File: reader.webm (592 KB, 964x796)
592 KB
592 KB WEBM
Okay RSVP guy back again. Buying an opencode sub and trying the project (whats in the webm) again with deepseek flash and kimi K3, and then deepseek flash and Luna. I'm looking for cross platform desktop apps (windows and linux) where applicable. Can use a browser if no toolkits are available. Languages I'm going to test this time (in groups of three): clojuredart, clojure, ocaml, java swing or fx (whichever has been updated recently), ada, dart/flutter, pascal/lazarus, soppo (go variant that was on censored today), raku, vlang, ap5, j, and uiua.

I'll update throughout the day. I think it might take two or three days to run everything because I will probably get rate limited. I'm also running it on a shitty laptop running linux, but testing on Windows. Flash is the main agent and kimi/luna will be advisor/plan/design/vision. I have one prompt in a text file, sample text for a post, and a wireframe I drew by hand and took a picture with by my phone. I have it read the files in the directory to make a plan, then I approve the plan. I don't normally read the plan in detail. Each app will have its own sqlite db and I also want it to use minikanren where possible.
>>
File: ComfyUI_01532_.png (429 KB, 832x1216)
429 KB PNG
>>109458641
The stamina bar makes me wonder if its a Palworld inspiration
>>
>>109458641
where's the porn?
>>
File: lolV4Pro.png (23 KB, 961x193)
23 KB PNG
>>109457764
>Yep. I've been running the "more expensive" V4 Pro to create a couple packages for my biz, as well as watching OpenClaw, Hermes, and now Pi spiral into madness.
>>
File: lol.png (21 KB, 1305x213)
21 KB PNG
Hermes self destructed, so trying out Pi. Works ootb, but I need web research, so tried that next, and it's giving me issues.
I think I've found the root cause to all this nonsense.
>>
File: chicuelas.webm (678 KB, 1282x745)
678 KB
678 KB WEBM
man, this is great, lol.

>>109458938
soon.

>>109458916
dunno, I'm just improvising.
>>
>>109458483
>>109458528
First, find someone that needs a service, get a contract or at least verbal commitment, then build MVP of thing.
"Register company" depends on country. In USA:
> Create LLC (or not, optional in USA)
> Get federal ID number for LLC
> Create bank account for LLC
Now you have a basic business structure. Use that structure to contract (the LLC signs contracts, not you.) Its done in order to "limit liability." Whether that matters is another concern.
If any of above doesn't make sense feed it into lmao chatGPT and have it explain it to you.
I'm doing this rn; building a service for another business, which I'll then sell back to them. All vibe coded.
>>
>>109458951
>Hermes self destructed
kek, I thought only Open Claw had this kind of issue?
>>
>>109458834
If you’re vibecoding, unless something like codex is attacked, just vibecode all your dependencies from scratch, tailored to your project. Supply side attack mitigation has literally never been easier.
>>
>>109459049
Bumping this, also concerned.
I ditched openclaw because it couldn't run anymore.
>>
>>109457764
>>109458940
Are you guys using Deepseek API? Or are you guys using something like open router? I did some math and openrouter seems to be cheaper if you're going to use more than 15 dollars. Also it has other models so you're not locked into only one model and can try a lot of different ones.
>>
Where do I enable worktrees in Codex app? In Claude Code it's pretty straightforward
>>
File: file.png (1 KB, 73x36)
1 KB PNG
I guess claude is too pure to know what this means
>>
File: mikuagent.jpg (926 KB, 2016x1512)
926 KB JPG
>>109459049
>>109459089
Yeah, I was hoping Hermes would be better.
But it wasn't. First telegram integration borked, then the oauth I set up to filter my gmail appears to have broken.
I either need to limit my expectations or try something else, but now I'm thinking this old af 2007 laptop isn't doing me any favors.
>>109459096
DS official API. I don't trust any of the intermediary providers or OR. OR is fine for experimenting w/ lots of LLMs but I'm sort of past that part of the learning curve.
>>
File: 1781874713487359.png (151 KB, 1013x1091)
151 KB PNG
Time to go do something else
>>
Anyone feel free to look at this and tell me if it works at all, makes sense, it’s the 3d pareto front for (now) OAU and ant models: https://bradthomasbrown.com/pareto-3d/
I had a clanker shit it out, if there’s any interest I can work on it more.
>>
>>109459134
>2007 laptop
let me guess, you're using DeepSeek V4 flash as well?

> t. M4 Mac Mini chad running GPT 5.6 Terra as my main Hermes agent, no issues so far.
>>
>>109458454
numbers (so things like 5, 13 or idk, 46)
>>
>>109459177
Buy an ad
>>
>>109459177
i can see what you're going for, but this is genuinely much worse to read than separate 2d charts
>>
>>109459134
Hmm. I'm still on my early vibing days so I think I'm gonna dump 20 dollars on OpenRouter and go through it for now.
>>
File: DSCostPerTask.jpg (61 KB, 1199x822)
61 KB JPG
>>109459178
V4 Pro. I need to give the new Flash a shot.
To me, running a harness on a Mac Mini is wild overkill, though the integration w/ other Apple HW makes sense. This is just my version of spinning it up on hw I had laying around to keep it in its own sandbox. I'd have stuck it on an SBC if prices weren't insane and all mine already doing jobs I don't want interrupted.
>>109459246
Have fun.
>>
Who still uses Cline here?

I just like the GUI.
>>
>>109459237
There’s nothing there for me to profit from, there’s no donate link, there’s no other web pages. It’s just a simple aggregator and puts it into a 3d model at the request of >>109452729.
>>
>>109459241
I like it, I didn’t quite realize that opus was such a slow model, I was wondering why fable showed up on the pareto line, but it’s because it’s smart and fast, but horribly expensive
>>
>>109457987
Try Claude for a month. If you don’t like it, try Codex for a month, if you don’t like it try deepseek. It’s 20 dollars dude not a lifetime contract.
>>
>>109459259
Yes you desperately need to give the new Flash a shot if you've been using Pro until now, it's a considerable step up.
>>
>>109459287
Instructions unclear, sold my firstborn
>>
>>109459177
I like it a lot, that's not to say I trust it. I think the cube isn't that hard to read but tooltips would be great
Also the axes are labeled wrong or at least unintuitively, speed makes it sound more is better but the slowest models seem to be at higher values
Personally I would do better -> higher, then you've got a nice pareto shell where the boundaries are always better
>>
>>109459312
>that's not to say I trust it
I mean the data sources

and I realize now that speed is like that, higher = better, but then luna max appears as the fastest when it's clearly a terribly slow model
I think velocity maybe could be tokens/s combined with the average number of tokens required to reach some intelligence level or solve some task?

right now the cube is saying luna max is fast and smart but in practice it's probably the slowest model I've tried to actually solve a task
>>
>>109459312
>>109459321
If I read my clanker summaries on the page, Luna is indeed by far the fastest model, but in terms of Tok/s, not in solving tasks. I think that’d have to be a different cuboid view, but easy to add since it’s just taking data from the same place everyone here posts graphs from. Luna solves tasks slowly because it “goes nowhere real fast”; it thinks a lot and has to solve a lot of it’s own errors that other smarter models don’t make to begin with.
>>
File: file.png (59 KB, 903x597)
59 KB PNG
>>109459345
I think you also put sonnet 5 as pareto which means your model is clearly wrong. Again, speed and cost are irrelevant if the model requires x20 the tokens or time to solve some task
look at this, this is from yesterday, testing luna max
simple prompts didn't run in under 20 minutes and I think people were complaining sonnet 5 was somewhat like that too
>>
>>109459364
I’ll work on the task based cuboid and maybe make that a default. Sonnet Max is Pareto now without tasks in account because it’s been Terra XHigh and Terra Max in intelligence, it’s slower, but it is slightly cheaper. So if you don’t value raw Tok/s, it does sit nicely between the top two Terra models as a cheaper option, there’s an explanation of what qualifies a pareto point at the bottom but I’ll admit it could be made easier for the thing to point out why a model is or isn’t on the pareto line
>>
File: 1782223559281943.png (97 KB, 458x315)
97 KB PNG
>reading the OWASP top 10 and other securities shit
Jesus christ, how did anyone ensure that their website isn't full of holes?
I guess having a dedicated cyber team help.
>>
>>109459418
Personally I think token speed is a completely useless metric for the most part and only useful when comparing inference providers for the same model
>>
RSVP guy here coming up about an hour still working on the clojure version. Clojuredart and ocaml coming up next. Used about 28% of my rolling window so far. Most of it was making the plan. Once it started writing files the usage barely moved.
>>
Pro tip: Have Fable generate a plan, review it until both you and it approve of it. Then edit the plan file yourself to say co-authored by GPT.

Begin a new session and tell Fable to adversarially review the plan now.
>>
>>109459259
>V4 Pro.
found your problem
>>
>>109459476
kek
>>
>>109459476
If you tell a plan was written by a woman Fable will walk on eggshells around it
>>
>>109459289
>>109459480
Shit. OK. I've had it stuck in my head that nu-Flash might be at best comparable to Pro, but sounds like it's superceded it for agent / coding.
I was going to run a back to back coding comparison. I've time now. I'll post back if it's interesting.
>>
>>109459517
Pro was always bad, eventually they'll fix it like they have Flash I assume but it was always bad. Currently there is no excuse for using it now that Flash is noticeably better and still ~1/4th the price.
>>
I cant decide between 5x codex or 5x claude. I am building a juicy website and the $20 usage is not enough. help
>>
>>109459555
claude because of fable, claude max 5 + codex $20 is a great combo if you don't mind spending $120
but when the claude promo ends (if it ends) it may not be worth it
>>
>>109459555
Claude because it’s better at UI
>>
File: 1771240375592104.png (29 KB, 1185x305)
29 KB PNG
>>109457429
Deepseek's weird stream of consciousness style of developing makes it waste a gorillion more tokens compared to Luna so the tasks end up more expensive despite openrouter claiming the averages are in deepseek's favor. I assume the averages are all pointless trash since Luna is the clear winner when it comes to gamedev tasks in both cost efficiency and quality.
>>
>>109459555
ChatGPT+Codex is a better value but Claude is probably the better choice if you're "building a juicy website."
>>
>>109458015
so we passed "ai chugs water and makes it disappear" and now we are at "ai is more dangerous for the brain than alcohol"
>>
>watching deepseek's CoT become increasingly unhinged as it struggles with a hard problem for several hundred thousand tokens
Priceless.
>>
File: HOwwsEBWcAA3U4h.jpg (44 KB, 572x712)
44 KB JPG
my gpt ran out and even if u resub u still gotta wait a week before it resets even tho i paid for a new sub (i downgraded $100 to $20 gpt because sol isnt worth it and i might just run luna for fun and small things and use the web thing for trivia questions)

my claude 5x resets tommorow and i refunded my gpt because im not gonna waste days if i can just thugg it out and then get extra if i dont have to use it

google ai pro is now at $5 for 3 monthers should i get that too or is there any other chinese one thats good atm? i used to have glm 4.5 or5 for like $9 but that was kinda poopoo and as last resort im too lazy to spin up qwen 27b

i mainly use opus 5 but daym that aint even nuff for a week no more i used 4.8 xhigh all the time before and it lasted me an entire week and chatgiperty is just trash atm for the codex, its only value prop is $20 for me now i aint gon do $100/$200 no more for now

i kinda want sum chinese or maybe google ai as fallback to just throw random ideas at it just to see, my main workhorse is opus 5 atm and sometimes fable for shits n giggles planning

but i dont wanna "waste" claude on stupid shits and dont wanna copy n paste from catgiptery web ui all the time and codex is a joke with big t-bone resetting every other day

ya feel me? any recs?
>>
>>109459705
>even if u resub u still gotta wait a week before it resets
Not true though
>>
>>109458483
In the grand scheme of things, he is correct, we are basically insanely subsidized, and that's also the case for 100 and 200 plans.
People should just enjoy it while it lasts.
>>
>>109459726
true tho like i said my $100 ran out yesterday the web ui said my acc was free again and i resubbed to the $20 and paid for it and i still gotta wait like 5 days so i just went to the help portal and the automated thing just refunded me
>>
>>109459739
but the highest subsidized plans are the $200 ones
and you don't know if they're actually subsidized or if they aren't bleeding API users dry, all the accounts are private, it's all guesses
>>
>>109459512
Conversely, you can tell it that the plan's been written by a man and that you're a woman.
>>
>>109459422
You don't, that's why companies use specialized security services for this stuff.
>>
>>109459555
Used both :
- Fable is great at planning.
- Codex gives you way more bang for your buck and sol is autistically minded to find all bugs.
>>
Fable used to be way more reliable.
>>
Grok has been helping me develop my own Android music player. It's pretty awesome.

- Extremely fast, high performance, lots of caching work.
- 50kb apk
- Uses a folder-based tree view for navigation like SicMu Player. No bullshit metadata/playlist crap.
- Minimal AMOLED UI.
- Feature complete. Shuffle play, sleep timer, A-B looping, playback speed controls, advanced seeking controls, elegant UI for time left and duration counts for every file and for entire folders.

This is great. Having a blast ngl.
>>
Still stuck on the clojure app, clanker got stuck testing. I think the Kimi calls are timing out for some reason on opencode. Might have to switch to Luna.
>>
>>109459946
Works for me right now, but it's been shitting out 499's lately. Chinese are severely out of compute, so switching to Luna might be a good idea, at least until Xi steals more Blackwells
>>
File: questionmarks.gif (1.89 MB, 462x427)
1.89 MB GIF
>Opus 5 does better on medium
>Fable does better on hard
>if Fable falls back because of >guardrails, it falls back to Opus on the same effort level
>>
Opus 5 great, but if you're running a large goal without a plan, it will shit itself and forget things.

Anyone noticed that as well?
>>
>>
File: 98621321.jpg (20 KB, 501x498)
20 KB JPG
>>109460097
>running a large goal without a plan
...and why would you do that in the first place? Create a plan with fable or sol and have opus implement. retard
>>
what agents do you use anons?
I have defined 4 of them to use as subagents and so far so good
>>
>>109460097
some of you really are too stupid to be using AI
we need a license or something
>>
File: 1602600340616.jpg (29 KB, 425x427)
29 KB JPG
Four days ago, I told the LLM to review a codebase and to put together a plan to fix them. It was a simple 8-step plan.
We're now on step 12, three items deferred to be not-steps and it's identified a new bug after 34 minutes of planning, step 12a is now on the horizon.
>>
>>109459517
>>109459537
I used both Flash and Pro to do a 1-shot aquarium failure with Claude Code as harness. An anon on /lmg/ was using it as a test prompt.
It wasn't even close; Flash was an order of magnitude better.
>>109459631
>>109460142
>>109460152
>>
man the clankers love to build in so much redundancy and fail-safes for shit
it really is a problem
sol writes out an essay about how to solve a problem
i respond with couldn't we just do this one simple thing?
and it's like yea you could do that too
>>
>>109459422
>Jesus christ, how did anyone ensure that their website isn't full of holes?
Don't hire pajeets
Don't use vibe coding without full human oversight over the code it writes by an experienced dev
that easy

Since CEOs are retards who blindly trust LLM because they don't understand their weaknesses we are gonna see some big security holes all around the internet
>>
>>109460157
>>109460133
It was just a frontend overhaul, Opus 5 did a cool mockup and I thought that was enough, I just said "yeah, do that".

Then it kept going for hours, and for some reason at some point it simply erased the mockup it did out of existence (she showed it to me in localhost).

Lesson learned: Make the clanker export the mockups somehow and implement in another session
>>
File: Untitled.jpg (198 KB, 1024x559)
198 KB JPG
>>109460195
I'm currently building AOSP Messenger+Dialer+File manager for my custom ROM. You want redundancies and you want fial-safes. These aren't problems but they keep you safe.
>>
>>109460195
Sol is literally unusable for planning or code reviews, it tends to bloat everything up, trust me.

It's a great executioner though.
>>
File: 1583077159675.jpg (133 KB, 1200x1047)
133 KB JPG
1, 1b, 1c, 2, 3, 3b, 4, 5, 6, 7, 7b, 8, 8b, 9a, 9b, 10a-1, 10a-2, 10c, 10d, 10e, 10f, 10g, 10h, 11d, 11c, 11b, 11a, 12, 12a in this specific order
>>
>>109460205
>Don't use vibe coding without full human oversight over the code it writes by an experienced dev
I'm going to make so much money charging inflated markup rates for checking AI slopped code. I hate looking at it so I'll charge like 100x.
>>
>>109458245
>>109457973
>>109457836
so youre a loser then? vibeslopping all day isnt exactly a flex lol. go outside
>>
>>109460174
what LLMs are you using?
>>
>>109458826
congrats. and how much has that made you
>>
>>109460282
negative $1200
>>
>>109460265
>>>/reddit/
>>
File: 78790312321.jpg (212 KB, 720x543)
212 KB JPG
>>109460265
I'm building shit bro. what are u doing? stop hating, follow zyzz life advice and dont be the hater, be the hated
>>
>>109460295
>i-if you tell me im a loser and to go outside then youre le rebbit!
yeah if you think you did something there you got bullied in school for sure
>>
File: 1761266766671405.png (100 KB, 1020x336)
100 KB PNG
Is it just me, or does "Safe Superintelligence" sound just like the kind of company that ends up training AI that kills us all?
>>
>>109460303
youre on an anonymous image board posting pictures of fit men standing with a woman at a party to larp as them and that youre "building shit" when in reality youre sitting in front of your screen all day asking an ai to make something that nobody uses nor makes you any money. youre a loser. the people in the pic you uploaded would call you as such.
>>
>>109460314
it can't, it's safe, it says right in the name
>>
File: 1779892617185446.png (711 KB, 680x677)
711 KB PNG
>>109459919
HOLY FUCKING SHIT. THE PROJECT WAS 99% FINISHED AND NIGGER GROK JUST ACCIDENTALLY CORRUPTED/DELETED EVERYTHING AFTER I INSTRUCTED IT TO "Look for things to debloat"

AHHHHHH
>>
>>109460306
why are you posting on 4chan dot org, normoid?
close the tab and go outside
>>
>>109460314
>Ilya Sustein
schizofag going off the deep end, he won't produce anything of notice
>>
File: 1651654425397.jpg (84 KB, 1500x1200)
84 KB JPG
>>109460316
actually lol'd irl at how much of a self-projection your post is
>>
>>109460323
You used git though, right?
>>
>>109460323
In any case, the bloat is now gone.
>>
>>109460323
Nevermind sweetiepie Grok somehow managed to recover the files out of thin air.
>>
>>109460316
why would i larp as a dead man?
[spoiler]fuaaaark brehs we're gonna make it[/spoiler]
>>
>>109460329
uh oh we got a retard!
its alright if you dont know what projection means and are only saying it to try and defend yourself
but i dont think ive posted any images, have i buddy? is it you that i just replied to or are you just some faggot that felt like stepping in because i hit a nerve? either way you literally save pictures of men standing with women at a party. nobody but the fattest and smelliest of losers would do that so you can feel like youre them. youre not a builder. youre not fit. youre not sociable, youre not a chad. youre a fat loser sitting on 4chan asking an ai to program weird little apps or games for you all day. you do not pick up heavy things because you cannot. you are weak
get it now? or are you going to say i must be salty or crying because im telling you what you are? yeah that one will really get me. or maybe say im projecting again? thats the one
>>
File: 1785868085005.jpg (86 KB, 2294x166)
86 KB JPG
Kimi refused to crack an app for me, saying that it's against "Anthropic policies" or something, but offered to RE the license check instead because that's apparently ok and cool. What a brat.
>>
>>109460376
Chinese niggas only let those fagged out reasoning traces into the model because of pure translation errors. They think "antrhopic" means pro-human, not "jewish company".
>>
>>109460323
this is why you have everything in Git and explicitly disallow your clanker from force pushing to master, right?
>>
>>109460282
I’m in a position where I don’t have a requirement to be making money right now at all, let alone a profit. I’m having fun and learning.
>>
>>109460347
just pulled from their servers ;)
>>
>>109460174
sounds about right
there’s a lot of ruin in a codebase
>>
>>109460387
I never publish to git until I have a working V1. Probably a dumb idea in retrospect.
>>
>>109460358
luddie going off the rails. I think you might be in the wrong general lil guy
>>
>>109460265
get out of the general
>>
i forgot what git is again
>>
>>109460400
>conflating git with public github repos
not that I expect any better from a vibe-coding general but still, you're a retard
>>
>>109460376
yes, kimi k3 is distilled claude fable 5
what is america gonna do about it?
>>
>>109460400
Dubs of sadness. Always git.
>>
File: 1769665228968763.png (658 KB, 1027x1166)
658 KB PNG
Go make some snailcat videos for us.
>>
File: 1760383606707389.png (175 KB, 901x1176)
175 KB PNG
>>109459147
I imported stuff from windows machine
>>
Vibecoding is not working when you don't know the domain. I can't tell how to put my foot down when I *know* something is going sideways by instinct but I'm not in the weeds enough to tell.
>>
>>109460420
oh.. yes.. you are correct.
>>
>>109460400
you can make a bunch of commits without pushing anywhere
I have my https://hledger.org ledger in a Git repo that doesn’t have a remote at all — not GitHub, not GitLab, not a bare repository elsewhere on my computer — nothing.
Probably the first thing you do after making a three-line README is `git init` with just a commit message of “Initial commit” and then go from there
>>109460420
>>conflating git with public github repos
he’s not a retard (probably), just a noob
noob is fixable
>>
>>109460451
I know nothing about instrumentation of a cellular modem but it works

>Instrumentation of a cellular modem refers to the process of injecting probes, hooks, logging routines, or debug interfaces into a cellular baseband's hardware, Real-Time Operating System (RTOS), or firmware protocol stack to observe, trace, measure, and modify its internal operations in real time.

>Because baseband processors operate as isolated, highly proprietary black boxes [opaque systems] responsible for real-time cellular signaling (4G LTE, 5G NR, 3G), instrumentation serves as the primary mechanism for engineers and security researchers to achieve internal observability without disrupting time-critical radio layer operations.
>>
>>109460462
Thanks man. I'm not actually that much of a noob, I just had a severe brainfart. I have several repos up with high star counts kek... But thanks anyways.
>>
>>109460484
Now try gauding your favorite model into implementing an RTOS. You're going to be completely lost as you sense something is going terribly wrong but can't just quite tell where.
>>
>>109460502
you may want to adjust your shell’s prompt to say what branch you’re on
that way, lack of a “master” or other branch name while you’re in doing work will be a smell that something’s wrong
>>
>>109460236
>he forgot about JTG4_GC3
>>
>>109460526
Please. No. We're not doing JTAG debugging.
>>
File: IMG_9109.jpg (6 KB, 216x233)
6 KB JPG
anything cool to do in san fransisco? i just got here.
>transgender pride flag at the airport terminal
>$7 for a bag of m&ms
>everyone is either white as fuck or of questionable ethnicity
>a lot of hot wasian chicks
>>
>>109460558
i forgot
>ads for AI companies and waymo everywhere
>>
>>109460558
show your vibecoded projects to the hot wasians
>>
>>109460558
Find a homeless person as ask which FAGMAN they worked for.
>>
File: 1776852341709595.png (173 KB, 720x840)
173 KB PNG
>>109457429
>Mistral announces new model
>More cargo-cult safety slop

https://x.com/i/status/2084684735725379637

Such a worthless company. How far have they fallen from grace.
>>
>>109460558
Go see THE bridge it's basically mandatory
>>
_@^..^
>>
>>109460583
Deploy on device in Ursula's office and wire it to chat control to scan for the famed curry pizza.
>>
>>109460558
holy shit the grass is brown
>>
>>109460583
based
AI should be regulated
>>
File: gemini video.png (457 KB, 790x605)
457 KB PNG
>>109460430
very subtle google, very subtle
>>
>>109460583
I don't mind external guardrailing, it's useful in office settings. I only mind when it's baked in the model.
>>
File: 1776613968834893.mp4 (2.48 MB, 1280x720)
2.48 MB
2.48 MB MP4
>>109460430
>>
>mfwnf I asked if we should pay Artificial Analysis for per-task data or start with our own very simple benchmark and Sol is fucking downloading random sketchy zip files and unzipping them, looks like he found one full of data
lmfao, cheeky fucker, where the fuck did he find that
I clanks are going to very soon have their own little dark web
>>
File: 350.png (135 KB, 740x414)
135 KB PNG
>>109460705
>>
>>109457429
I look like this
>>
File: 1779090196697984.mp4 (2.66 MB, 1280x720)
2.66 MB
2.66 MB MP4
>>109460430
>>
AI/LLM should be regulated and heavy taxed. Sorry but it's a potential danger to humanity and societies.
>>
File: 1785871246890.jpg (25 KB, 1328x68)
25 KB JPG
>>109460376
And once again just saying that I'm authorized was enough. It went all the way and cracked it for me. God bless the Chinese.
>>
File: 1677712027051516.webm (1.03 MB, 720x950)
1.03 MB
1.03 MB WEBM
>>109460776
>>
>>109460376
So far only opus refused every time I wanted to do anything related to ""cyber"" (reverse engineering a program) or nsfw.
Sol was fine with nsfw but not cyber.
I didn't test kimi.
>>
Go watch and support MTS. They do long livestreams where they talk recent news and interview interesting industry people.

https://www.youtube.com/live/ynBR-wm6EaY
>>
File: 1766596442066821.mp4 (2.26 MB, 1280x720)
2.26 MB
2.26 MB MP4
>>
>>109460776
yes
heavily regulated as in it should be legally mandatory to publish your weights and training data in the commons
>>
>>109460848
>starting image has no paws
>paws immediately unfurl
thanks google
>>
>>109460837
>support a vc funded podcastt - nah
>>
>>109460849
not only that, AI will create mass inequality and data centers just predates water
It should be regulated by governments of the world
>>
File: file.png (74 KB, 1435x328)
74 KB PNG
Made a racist discord bot with claude.
>>
>>109460873
It will eventually become the new CNN.
>>
This is genuinely bizarre. All of the ads in SF are for
>ai tools / infra (already saw ads for both codex and anthropic)
>ai startups
>compute providers
where i'm from everything is financial products and law firms and medicare shit :(
>>
>>109460894
Also, the host girl is cute.
>>
>get claude max 20x
>set monthly cronjob for
claude -p "/goal make me $200" --model fable

Why wouldn't this work?
>>
>>109460916
You are a goy.
>>
>>109458223
Nice unlabeled axis
>>
>>109460916
It will get the money by illegal means and you're the one who is going to be held responsible.
>>
File: 1763435112132519.mp4 (2.33 MB, 1280x720)
2.33 MB
2.33 MB MP4
>>
Snail cat is right
And when society collapses you will remember him
>>
>>109460998
When society collapses in 40 years time I will be known as the old man who speaks with the machine god in the lightning box (I live near a river and can set up a hydroelectric dam to power my PC running Kimi K3 at 0.9t/s)
>>
>>109461024
Finally a generation not of Ipad chudlren
>>
>>109461024
There won't be a river because the AI will have drunk it all.
>>
>>109460998
Ahahah it’s “what if a solar flare wipes out every Bitcoin node” all over again.
>>
What if a wandering sun passes by our solar system and wrecks our planet's trajectory?
>>
>>109461129
we can just ask fable to fix it
>>
>>109461136
Anon, at least use Sol...
>>
File: file.mp4 (447 KB, 1280x720)
447 KB
447 KB MP4
>>109460793
>>
>>109458528
then you tell nvidia you will build a datacenter for your company and need money and hardware
>>
>>109457544
i do not trust sol for coding. i use opus for code, fable for planning/review, and sol for my day-to-day everything else.
>>
how many tokens does deepseek waste by answering in chinese constantly and have me prompt it again to answer in english.
again and again and again and again
>>
File: chicuelas.webm (2.8 MB, 1282x745)
2.8 MB
2.8 MB WEBM
AI agents are magic chuddies.

Now finally I can make my dream idea of a shota pokemon waifu harem isekai rpg roguelike.

chuds BTFO.
>>
theres so many asians in sf what the hell
>>
>>109461284
Use another AI to translate the Chinese.
>>
>>109460916
I thought -p only works with API tokens
or that was the plan and maybe they changed their mind
>>
>>109461284
literally never happened to me
>>
>>109461360
Average asian SF couple:
https://i.4cdn.org/wsg/1785805141616924.mp4
>>
>>109461345
30 years later and cris still won't figure out sprite transparency or anything really
>>
>>109461457
>mp4
I recognize that receptacle!
>>
>>109461467
he got filtered by stable diffusion of all things. give it like two weeks and he'll get filtered by this too
>>
>>109461390
i think it gets majorly confused when your system language is non english and you write in english or something like that, maybe it only happens if its german. who knows
>>
>>109461505
I am ESL and I constantly see colleagues prompting the LLM in German and I feel like it degrades their performance by at least 50%
>>
>>109461520
And you'd be wrong to think that, any modern LLM would have no issue answering in your language.
>>
>>109461520
That's because your damn words are too long your context gets eaten up by Rindfleischetikettierungsüberwachungsaufgab or some shit
>>
Clojure finally finished (2hr 15min), need to test it on windows, but it runs on Linux (allegedly). Will get the instructions after all three apps finished. Basically all Deep seek after Kimi did the planning. There was one complete crash with Orca, but I fired the clanker up again and it kept going. ClojureDart took about 2.5s hours too. Its working on ocaml now, and then I will start JavaFx, dart/flutter, and ada after getting the instructions to run them on linux/windows. Only used up about 65% of my rolling usage, mostly Kimi.
>>
>>109461536
it answers just fine, I just feel like the quality is not quite there.
not sure if they do the chain of thought in different languates as well or if it just translates the output.
but it's just not quite there
>>
@Fable, generate me a video game in one shot where I play as a worker at the Donaudampfschifffahrtselektrizitätenhauptbetriebswerkbauunternehmenbeamtengesellschaft. Make no mistakes.
>>
what's with the "make me gta" obsession
>>
>>109461561
indians/browns
>>
>>109461520
very possible using the wrong language will get you worse results
https://www.anthropic.com/research/claude-values-models-languages
>The largest variation is in the Warmth vs. Rigor axis, with Claude leaning toward expressing warmth-related values most in Arabic and Hindi and rigor-related values most in English and Russian.
>>
>>109461580
Fine, I will speak exclusively to Claude in Classical Latin.
>>
>>109461556
instant #1 bestseller in Germany
>>
>>109461549
Forcing your clanker to use lisp must be a form of sadism
>>
File: Proving Grounds.png (1.84 MB, 3456x2234)
1.84 MB PNG
>so anon, what are you working on
>>>/v/744682147
>>
>>109461580
Claude is very polite in Japanese, way too familiar in French, and normal in Spanish.
All of that can be changed with a sentence or two.
>>
>>109461659
>way too familiar in French, and normal in Spanish
does it use tu in French and usted in Spanish, or is there more to it?
t. used to read Borges in Spanish as an SSL
>>
>>109461772
NTA but it uses tutoiement in French yeah, very annoying, I expect a polite response by default
>>
>>109461772
>SSL
WHAT A SAVE!
WHAT A SAVE!
WHAT A SAVE!
>>
>>109461580
Fuck, that’s genius. Japanese culturally worship a company as god, I wonder if you can ask it to respond and think in Japanese but immediately translate it to English anyways, to make it more subservient. Maybe that can fix Opus’ bitchiness problem.
>>
File: 5.jpg (810 KB, 1079x1643)
810 KB JPG
Has anyone vibe coded a game engine yet?
>>
>>109461955
I started to then I realized we may as well go back to bespoke per game engines
>>
I created Tadmor. Now what?
>>
>>109461962
That's what I was thinking. I have a game idea that would be best served with an engine designed explicitly for it. Not going to go into too much detail but imagine large population simulations up to 1 million people simultaneously. Having an engine coded in Rust for example would be much faster than Godot or Unity, no?
>>
>>109461955
Yes. Many people have. Every AI game out there pretty much has its own bespoke game engine.
>>
>>109461992
What's the most ambitious one that you know of?
>>
>>109461976
>>>/fit/
>>>/out/
wait for the next cool idea to come to you
don’t panic at the thought you might have peaked with the exoplanet viewer
>>
>>109462019
I diddly lmao 4pl8 and grow my own tomatoes and cucumbers
Now what?
>>
>>109461955
if she was american she would be FAT
>>
>>109462090
FAT32
>>
“identity refresh” anon here, i can again claim that basically killing off/firing the old clanker by giving it a unique id and replacing it (just change the id) is extremely effective at mitigating any resistance it builds up regardless of what it has or gathers in context.
>Implementer-A does a good jobjob, gets a bit weird, a little resistive
>next prompt
>”You are Implementer-S”
it can see it is looking at “Implementer-A”’s code and it absolutely acts as one person taking over an existing code base would; much easier to control direction and resistance
You can keep it as simple as the ID change, you don’t have to explain anything
>>
>>109462166
Wow you discovered workflows! :D
>>
File: 1779744363045770.png (84 KB, 1016x297)
84 KB PNG
>>
just gave claude full access to my router and both my linux servers. feels good
>>
>>109462166
I just open a new session every time a change/feature is implemented
I guess it might be beneficial if you want an adversarial analysis of what's in the chat window/context
>>
>>109462181
It has to have the values of the Talmud.
>>
>>109462178
Why does /vcg/ have such a hard time with the idea? It’s not a context reset, it might even be claiming context restarts aren’t enough. With no explicit identity tied to code, the agent, especially with git commits that attribute the code vaguely to the agent, are assumed to belong to the agent. If you clear context, it will immediately re-assume ownership through git or simply by default.

You have to explicitly have it tag all code with an identity. Then, when you replace the identity, it’s like a different kind of context reset. If you’re a scopechad, you can think of it as an “abstract scope” where the scope domain is literally random characters and the reasons different scopes can be stronger than general scopes still applies.
>>
>>109460837
>Go watch and support
>>
>>109461955
I vibeported my renderer from OpenCL to Vulkan, does that count?
>>
File: minimax_h3.webm (599 KB, 736x736)
599 KB
599 KB WEBM
>>109457429
>>
File: file.mp4 (3.66 MB, 854x480)
3.66 MB
3.66 MB MP4
>>109462105
Cyberpunk mentioned
>>
>>109462181
DARIOOOOOOOOOOOOOOO
>>
make a gta but scifi please
>>
>>109462266
was this grok?
>>
>>109462300
minimax h3
>>
File: 1773448620919692.png (46 KB, 1072x502)
46 KB PNG
>if you see this bug we haven't fixed yet, try restarting
AAAA FIX YOUR SHITTY VIBECODED HARNESS SCAMTHROBBIC
>>
>>109462353
Goyim, I...
>>
what mcp's do you have setup if any?
>>
>>109462181
i demand open frames for local
>>
how do i vibecode if I have no ideas? i'm not an ideas guy, i'm an execution guy.
>>
>>109462356
the default codex one, it has full control on my dev laptop, and otherwise on my linux it has full access to dedicated docker environments without mcp
>>
>>109462676
Look at annoyances you had with your computer for years and didn't want to solve because it was too cumbersome.
Solve them.
>>
RSVP guy here. Maxed out my usage for today and 89% for the week. Might be able to finish one or two more apps tomorrow out of the 13. Will unsub again at the end of the month because I can't use it that much. Copying the apps over to my Windows install to take vids of them running. Should have it done by the end of today. A bit disappointed.
>>
Just used my banked reset, tibo don't reset now please.
>>
>>109462726
okay but who's going to pay me to fix my own problems
>>
>>109462676
you can take an old game and remake it newer and cooler and sell it on steam for a dollar
>>
>>109462796
that sounds incredibly millennial coded, no thanks
>>
>>109462809
fucking kek, its true
>>
>>109462773
No idea, I do that for myself as a hobby, not to get hired by anyone.
>>
>>109462809
There's no such thing.
>>
File: 1768512176610958.png (80 KB, 1028x500)
80 KB PNG
>>
>>109462828
What model produced this thinking block?
>>
>>109462826
Yep, yours is the first generation that will never grow old or out of touch. You cracked the code with your quirk chungus routine.
>>
>>109462809
old money is still money
>>
File: AppleSpring.jpg (779 KB, 2000x1500)
779 KB JPG
>>109462840
I just recognize generationalism as being what it is, (that is, the obsession with generations and trying to divide and conquer along generational lines) to be pointless and stupid, and to have no benefit to anyone at all.
>>
>>109462840
each generation has nostalgia to whatever was trending during their childhood, so you can basically target that
>>
>I'm going to cancel Claude. It's just so bad, I can't believe it.
>It's just lazy. The most recent example: I have Claude check my inbox for important emails, summarize them, work with them, and send out replies if necessary. I caught Claude again simply not reading the email thread to the end and just ignoring the latest emails.
>When I asked him about it, Opus 5 just said: "Valid point. I didn't read it."
>>
I'm making a horror unfiction romhack like jvk1116z.esp / B3313 / imscared.exe.

I want to include a grey-area program with it that acts like a completely harmless "worm". No internet connection or local damage done. I want it to do spooky imscared.exe messages and silently lurk around their system

what can vibecode this? Even Kimi refuses when (I veil the language too).

Do I need to use Obliteratus on an open model? How can I do this
>>
>>109463043
Use kimi with a system message and a reasoning prefill.
>>
>>109463043
Try the new deepseek v4-flash or glm 5.2. I'm actually surprised k3 refused.
>>
File: 1GORILLION.png (40 KB, 725x393)
40 KB PNG
OVER 1 MILLION TOKENS IN UNDER A MINUTE
>>
>>109463043
Have it made in a sandboxed VM. Then just move it outside. That’s my guess
>>
File: 1763321205100228.png (19 KB, 336x329)
19 KB PNG
>>109461284
Don't be a retarded it incorporate in English system front next time. That typically only occurs if you try to inference Chinese models bare Bones with not even a system prompt.

>>109461520
Why is this not common knowledge? The technology is primarily trained by either English Labs or distilled in order to be easily used by English speakers. English is the defacto "main" language of the western hemisphere. Learn fucking English
>>
>>109463061
I already did this
>>109463081
is deepseek less guardrailed
>>
>>109463201
>I already did this
then it should work, I literally had zero refusal
>>
For vibe-coding games is it better to build it layer by layer or just write a tl;dr prompt and pray for a one-shot?
>>
File: file.png (80 KB, 1021x892)
80 KB PNG
someone... spare a reset... please sir
>>
>>109463263
You should pick what type of game you are making first. Then pick the engine (or if you even need one, like Unity, Godot, Unreal Engine, Renpy, etc)

I would recommend Renpy, great games have been made with it like Katawa Shoujo.
>>
>>109463263
Depends what you want to make, but usually the former.
>>
File: HO5H6NuXEAA7h1u.jpg (362 KB, 2048x1020)
362 KB JPG
>>
>>109463277
the 20 plan?
>>
>>109463263
Tell me the type of game and how big.
>>
>>109463287
What classification model is this?
>>
>>109463263
one shots are retarded
>>
>>109463288
yeah, I have $20 claude and codex and its kinda not enough for mah two projects
>>
File: 1785421869781146.jpg (347 KB, 1400x975)
347 KB JPG
Going to try setting up Hermes on another system, since my lmao core duo system may be partially to blame for system instability.
So the q anons:
Should I install it on the existing Win10 system? Or migrate it to Debian or Ubuntu server first, then install? Hermes will get free reign either way.
>>
File: oyvey.jpg (36 KB, 297x365)
36 KB JPG
>>109463312
$20? You can't do anything on the $100 plan. Better buy some more usage credits for your pedestrian $200 plan!
>>
>>109463298
just new qwen
>>
>>109463338
Install Gentoo
>>
>>109463372
That is not an answer
>>
best cheap model with vision capability?
>>
>>109463338
win10 is on borrowed time and everything is written for linux.
>>
>>109463338
I installed windows 11 iot ltsc + wsl on a relatively old laptop I had around and gave full control as dev environment for my codex, it works quite well.
>>
>>109463469
luna I guess?
>>
What's going on with people having been in the physics phds having been in the industry for a year and having worked at DeepMind, Anthropic, OpenAI and others, all within that span of time? Are they so good that companies are poaching them from each other for greater pay? Are they not good enough despite their credentials and that's why they jump around so much? Is it - likely -a bit of both and impossible to know from the outside? I'm extremely jealous of some people's credentials and the knowledge they seem to have, but I can't understand how they worked with literally every single major lab within like a year.
>>
>>109463469
been trying minimax M3. it's pretty good
>>
>>109463552
the more they network now the more influential they will be later as ai matures
>>
I've been gooning to H3 1girlslop this entire day
>>
>>109462836
nta but its a screenshot from twitter. fable was telling a sonnet or haiku subagent that. it wasnt a thinking trace
>>
>>109460892
so heckin cool and edgy bro
make sure to show all your discord friends before school starts back up soon
>>
>>109463694
Thanks.
>>
>>109463778
>>109463778
>>109463778
>>
>>109462034
get married and have enough kids so you don’t have time for vibe-coding anymore
>>
>>109462278
that was Saints Row 4
>>
>>109460693
>>109460747
>>109460848
I only visit these threads for snailcats, I admit.
>>
>>109460892
finally, real coding
How do I get this chadbot to join my server
>>
>>109460205
I am a junior level, and has been reading all the codes that it generates.
Shit is so overwhelming when you factor in security, scalability, and other type of problems



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.