>>109567558
Okay, but uh, I need a retard tax for this: where's the "make lunar agent go do this" button in Codex?

What are you vibe-coding anon?

File: file.png (2.07 MB, 1448x1086)

2.07 MB PNG

**>August 14th - Z.ai drops GLM-5.3**
Same base as 5.2 + heavy post-training. Claims strongest open coding model (massive jumps on Terminal-Bench 3.0, DeepSWE, etc.), agentic performance approaching Fable 5 in places, and standout cyber capabilities (top-tier CyberGym scores, already finding real vulns including reports involving Cursor). Live via Coding Plan/ZCode. Open weights held ~2 weeks for safety hardening.

**>August 14th - Qwen3.8-27B open weights ship (Apache 2.0)**
After the delay drama, the practical 27B multimodal dense model is out (262K native context, extendable to 1M, strong coding/office/agent claims). Max-level 2.4T-A95B weights also confirmed available. Both on Hugging Face + ModelScope.

**>August 13th - DeepSeek V4-Pro-0813 leaves preview**
Official flagship out. Vendor numbers solid in agent/cyber spots, but independent benches underwhelmed vs hype + big price hikes incoming. Mixed reception.

**>August 13th - Google Gemini 3.7 Flash**
Workhorse coding/agent upgrade over 3.6 Flash with notable benchmark gains, lower pricing.

**>August 12th - SpaceXAI Grok 4.6**
Ties GPT-5.6 Sol on Artificial Analysis Intelligence Index. Same $2/$6 pricing. Long-running agents + coding focus.

**>August 11th - NVIDIA Nemotron 3.5 Lightning (open)**
30B MoE / 3B active for fast high-volume agent execution.

**>August 10th - Meta Muse Glimmer (open)**
30B Apache 2.0 local single-GPU agentic model.

No major new closed frontier flagships today. Side noise includes continued GLM-5.3 cyber discussion, Anthropic watermarking rollout, and enterprise/revenue chatter.

>>109567587
Why do we love indian women?

dipsy is so goated. sorry for doubting you Xi

>>109567641
We don't, the OP is a schizo who deliberately posts the worst version of the OP paired with an image of piss-and-shit Miku as a form of trolling.

File: file.png (11 KB, 419x644)

11 KB PNG

>>109567602
oh no it's terrarizing me

File: IMG_2579.jpg (192 KB, 534x528)

192 KB JPG

>>109567587
Fuck off jeet
This is a vibeGOD racist thread

I'm using cursor to change a roguelike since I have the source code
Holy shit this just works. It's amazing. It's wonderful.
It's doing basic black bitch levels of changing but it's amazing

>>109567708
you high anon

File: chicuelas.webm (1.39 MB, 1278x748)

1.39 MB WEBM

had to retard wrangle deepseek by drawing a mask, but my autotiles now work great.

Both autotiles and the tile sprites were made with AI.
Chat gpt made the first autotile, then converted it into a real autotile with aseprite and tilesetter.

>>109567587
I don't use git—I use AI.

>>109567838
based, why learn git if you can just ask the AI do undo shit?

vibe coding status on black communities?

>>109567587
Lol reminds me of the Indian Gemma personification from months ago.

File: kaggri.png (199 KB, 958x881)

199 KB PNG

I think I got the RL loop to finally work although I will be closely monitoring for a while.
But once I confirm it works I'll focus on optimizing the code.
The policy loss jump at the end is because the model is composed of two parts and I had forgotten to unfreeze one of them and apparently it also applied dropout which messed up with it. Although this is unconfirmed, we'll see if it improves after "fixing it".

File: value_training1.png (346 KB, 1680x1890)

346 KB PNG

Nevermind wrong image, this is the latest one.

File: Screenshot_20260815_00460(...).jpg (646 KB, 1079x1190)

646 KB JPG

>be me, millennial, uptime longer than any zoomer
>zoomer flexes vibe coding circles around u with claude
>i git clone`his masterpiece, run valgrind 890MB leaks, 12-use-after free, 4 segfaults just on init
>strace -c his prod — 2,400 syscalls per GET, kernel context switching harder than his gender
>perf stat shows 97% CPU wasted on Ai generated recursion and require() orgies
>objdump his binary — 400MB, includes 3 copies of lodash and a full react devtools
>tcpdump -i any his app phones home to 47 different AI telemetry endpoints per request
> grep -c "error" /var/log/his_shit` —9,000 vs 12 "success", all hallucinated routes 404ing
>zoomer sees my terminal, pastes the entire backtrace into cursor.sh
>AI hallucinates a fix sudo rm -rf /tmp/* kills his own session cache, site goes 500
>i type 4 lines of ebpf, one sed, and a raw sendto() socket
>his entire vibe runs 0.004ms, traffic rerouted, his AI agents all hit rate limits
>2 c4n pl4y? b1tch, ur A1 wr1t3s sp4gh3tt1 — 1 wr1t3 th3 f0rk() th4t ch3ws 1t up`
>u run circles? i memcpy() over them while ur box chokes on cache misses
>skid ur vibe ust got SIGKILLd by ^<>^

>This went under the radar this week, but we just shipped the ability for models with multi agents v2 to delegate to any supported model, including Luna!

>>109568069
i posted it earlier. its a shame i have literally nothing to vibecode right now

Anyone know how to jailbreak claude code into doing shit that its not supposed to? I want it to make it study a pirate stream site so I can make an application using its API but the gotcha is these sites using shady ways of concealing their origin.

>>109568003
my biggest accomplishment in life was replacing an xinit'd window manager in real time by attaching to it with a debugger and execveing the other window manager's executable
I peaked there I will never be able to surpass that

File: 1766341900411595.png (81 KB, 208x243)

81 KB PNG

WHY THE FUCK DO I HAVE TO PUT THE ; AFTER EVERY SINGLE LINE OF RUST CODE WHAT THE FUCKKKKK

File: file.png (7 KB, 374x114)

7 KB PNG

I'm just a cheerleader

>>109568105
>t. Python baby

>>109568100
that ironic post is the closest you've ever come to actual engineering, and even then you had to invent it as a joke because reality gave you nothing but npm install failures and AI hallucinations. you've never attached a debugger to anything except your own self doubt, and the only execve() you've ever seen is when you executed your last shred of credibility by replacing real skills with a chat window, kek, we'll be over here, still writing the code your AI can't even understand, while you're busy being ironic about a life you've never had.

>>109568105
cobol wins again.

>>109568153
Uhh no I've actually done that.
I also compiled Mosaic for a modern (in ~2011) version of Debian, made a toy operating system (pre AI) in asm and C that booted to protected mode and drew characters you typed with the keyboard in graphic mode.
Made a primitive physics engine in C for a flight simulator that I never completed.
Did lots of small scripts and such.
Did some webdev.
Played around with AIML chatbots and some small amount of ML experiments with MLPs and RNNs pre-transformers.
Played around with Metamath, automated theorem provers like Prover9 and Vampire an>>109568153
Uhh no I've actually done that.
I also compiled Mosaic for a modern (in ~2011) version of Debian, made a toy operating system (pre AI) in asm and C that booted to protected mode and drew characters you typed with the keyboard in graphic mode.
Made a primitive physics engine in C for a flight simulator that I never completed.
Did lots of small scripts and such.
Did some webdev.
Played around with AIML chatbots and some small amount of ML experiments with MLPs and RNNs pre-transformers.
Played around with Metamath, automated theorem provers like Prover9 and Vampire and non Turing complete languages like Charity.
Competed in other anons in programming competitions like this one https://desuarchive.org/g/thread/67984268
Made a YOLO based captcha solver and anon simulator with the base version of Llama 2.
Did some (unsuccessful) algorithmic trading and other minor projects.
But no big projects until AI, which has made programming much more fun for me tbqh.d non Turing complete languages like Charity.
Competed in other anons in programming competitions like this one https://desuarchive.org/g/thread/67984268
Made a YOLO based captcha solver and anon simulator with the base version of Llama 2.
Did some (unsuccessful) algorithmic trading and other minor projects.
But no big projects until AI, which has made programming much more fun for me tbqh.

File: 1760821914436217.png (12 KB, 820x521)

12 KB PNG

holy shit we're vibe coding

File: KEK.png (28 KB, 1234x108)

28 KB PNG

> meanwhile, at Anthropic
> pic related
kek, at this rate even Google will surpass them soon

On MacOS, is it better to use Osaurus over Hermes Agent?

File: Frontend.png (401 KB, 1911x1642)

401 KB PNG

>>109567587
reposting from >>109567364
I made a TTS app (opensource) that wraps 41 engines. Most repos cover a handful at best, and I got sick of remembering which tool did which voice well and jumping between them per project. Main difference is it's character-first, not engine-first: you save a character, build a library, and plug them into whatever engine suits the job.
►It has:
>auto voice binding for zero shot: feed it a file/dataset for a character and it binds across every available zeroshot engine and makes test samples, so you can hear which fits and set that engine as the default for that character
>API access for hooking it into other tools
>real-time voices with cloning for chatbots, plus decent presets
>diarizer: point it at a folder with a show/anime/whatever, it cleans it, splits by speaker (80-90%), works out who's talking and hands you samples. Works better than I expected, still a constant WIP
>full BTVA DB + AniList with database updating, if I can share it. Makes matching characters to VAs trivial, or finding their other roles ("Troy Baker as X is a thin sample, what else has he done?")
►Guts are there but not fleshed out/tested:
>podcast creation
>book conversion
>emotional diarization
>speech to speech (some STS engines already in, want TTS nailed first)
>TTS/STS/RVC lora/model trainer
>live RVC (Okada but better)
>RVC v2
►Ideas from feedback in the vibeslop general:
>browser extension for right click TTS on webpages
>live translation
Holding the ebook/podcast converter back until I work out which engines handle long form, otherwise I'll be tweaking forever and never post it. The best sounding engines that nail emotion tend to drift quick, so each needs testing before it goes in the guide.
►What I want to know:
Would you use it, is it a shit idea or not, and what would you want added? My friends aren't into TTS and won't shit on it hard enough to catch something obvious.

File: a6d204fe90.webm (3.35 MB, 492x300)

3.35 MB WEBM

Some anon made this game a few weeks ago, but he disappeared.

File: 1781148601143136.jpg (683 KB, 1080x1054)

683 KB JPG

>>109568239
I can't sincerely believe you vibecoded a fucking hello world jesus fucking christ
Humanity is fucking doomed if we go down that path, I see it now

File: value_training2.png (350 KB, 1680x1890)

350 KB PNG

>>109567929
Run saved!

>>109568239
opus 5 ultracode so goated...

>>109568239
Fucking based well done and welcome to the future

end of the line, claude pro sub ends today and i ran out of weekly usage anyway. not liking watermarks or reduced usage
havent tried codex or paid for gpt anything because chatgpt web free scared me off in 2023 and claude free was significantly more better and so i ended up starting to pay for claude pro 2 years later
how's it going on the openai side? i personally thought opus 3, then 4, then 4.1 and so on was pretty good but i was using sonnet the entire time until recently with opus 4.5 where i've been doing nothing but using opus on claude code since

>>109568258
>Osaurus
who?

>>109568389
Sol is an autistic savant. It depends on what work you like doing. I still think it makes the worst possible decisions when it comes to anything related to human interactions but is extremely sharp when it comes to anything technical
If I have a bug or a technical issue I can just ask sol to find it and he will. I don’t think there was even a single bug or issue I had that sol couldn’t fix. I’m still trying to understand the technical ceiling of Sol
Apparently luna is a very good and cheap implementer at xhigh or max but pretty slow

>>109568408
It's right there in the OP you fucking moron.

>>109568239
you're all laughing, but this anon will deliver GTA 6 before Rockstar

>>109567855
>assuming your clanker remembers exactly what was done
one nice thing about using Git (regardless of whether you upload things to a forge like GitHub or whatever) is that you can talk to your clanker about changes made on such-and-such date
because git commits have datetimestamps

>>109568105
welcome to programming! Pretty ambitious to start out with Rust as your first language!

>>109568258
a few questions:
> Why do you need something like Hermes?
> What model do you plan to use as your agent?

>>109568239
welcome to the party

>>109568213
>t. AI