/aicg/ - A general dedicated to the discussion and development of AI chatbots.GLM Edition>The BotBrowser extension for SillyTavern has been confirmed to be a trojan that steals API keys. You should remove the extension and consider rotating your API keys if you've used it. Details:https://rentry.co/st-backdoorThe StructuredPrefill extension by the same author is suspected to be compromised too, so using it is not recommended either.>NewsZ.ai releases GLM 5.2: https://z.ai/blog/glm-5.2MiniMaxAI releases MiniMax-M3: https://huggingface.co/MiniMaxAI/MiniMax-M3Moonshoot releases Kimi 2.7-code:https://huggingface.co/moonshotai/Kimi-K2.7-CodeUS government unreleases Claude Mythos 5 and Claude Fable 5: https://x.com/AnthropicAI/status/2065597531644743999Anthropic releases Claude Mythos 5 and Claude Fable 5: https://www.anthropic.com/news/claude-fable-5-mythos-5Nex AGI releases Nex-N2-Pro and Nex-N2-Mini: https://huggingface.co/nex-agi/Nex-N2-ProNVIDIA releases Nemotron 3 Ultra: https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/>FrontendsSillyTavern: https://docs.sillytavern.appRisuAI: https://risuai.netAgnai: https://agnai.chat | https://rentry.org/agnai_guides_>Botshttps://chub.aihttps://realm.risuai.nethttps://char-archive.evulid.cc/shutdown.htmlhttps://chararc.bernkastel.pictures/https://archive.cardbox.moe/https://botbooru.com/>ModelsJailbreaks: https://rentry.org/jb-listingGPT: https://platform.openai.com/docsClaude: https://docs.anthropic.com | https://rentry.org/how2claudeGemini: https://ai.google.dev/docs | https://rentry.org/gemini-qrDeepSeek: https://api-docs.deepseek.comGrok: https://docs.x.ai/overviewLocal: >>>/g/lmg | https://huggingface.co/models | https://openrouter.ai>MetaLore: https://rentry.org/aicg_chroniclesLog reader: https://sprites.neocities.org/l/r>Previous Thread>>109076349
desu
A N C H O R
In other news, Midjourney announces that instead of releasing a new image model, they're building a machine that makes taking a detailed 3D scan of your rectum easier. They're also building a spa.
>>109081389Go on a brainstorming session with AI. It's actually pretty fun to get your thoughts out that way.
>>109081693I thought standard protocol was to chat about your idea with a low model and then engage the higher model to make the card
>>109081704Idk about "standard" but I go to my character creator card and discuss my partially formed idea with opus 4.6 or deepseek v4. We throw some ideas around, it offers some suggestions, I mix and match and think until I'm satisfied then I go section by section and have AI write me a draft for proofreading. Deepseek v4 was previous the best at writing but I think that honor now goes to GLM 5.2. Gemma 4 is the best at writing succinctly so after you've put all the sections together you can have Gemma trim it down and proof read that. Done.
>>109081716My workflow is more like feeding wikis and reference images to gpt-5.5 on medium and then flip it to xhigh to create a SillyTavern .png
>>109081729>more like feeding wikisI mean yeah if it's not an OC that works too. I haven't even touched gpt since 4.5. It doesn't filter you at all?
the nice thing with the fable shitshow is that it should vaccinate at least for a while most other companies from playing the "our model is so dangerous it's like nuclear weapons" bullshit
>>109081733I haven't run into anything filterable yet.
>>109081737My fear is that everyone will become safetycucks now, especially local releases. I have no faith in the west at all.
There's no official API i can pay for and plug into ST like with deepseek?
>>109081747yeah that's a huge issue but frankly speaking : - anthropic was full on like that anyway- gpt has been unusable for nsfw- gemini is already like that even if you disable all guardrails, it still gives you more shit in 3.1 vs 2.5I can only hope the next finetune of deepseek will be good enough for rp/story
>>109081756What? You can pay for all of them.
>>109081756open router, plenty models to choose from, payg
>オムレツ
>>109081834>ライス
AIEEEEE I JUST WANT TO USE CLAUDE WHY IS IT SO EXPENSIVE FUCK DEEPSEEK 4 IS SO FUCKING SOULLESS
>>109081623I am using AI to master linux. I've never used it before and I'm really bad at it but I'm improving. The thing is, I'm so new I don't really have any ideas for projects to help me learn further. Any recommendations for prompts to use to learn more abut linux? Like a specific project I could try building
>>109081846Consider employment>>109081849Set up a Hetzner VPS
>>109081846try glm? the new one is pretty good apparently.
>>109081846Deepseek is actually very soulful, it's just passive and positivity slopped.
>>109081851I can pay, it's just that paying 20 bucks for a single rp session is insanity, there has to be way to get something pretty good for half of that
>>109081865Holy fuck how long are your RP sessions? Do you use prompt caching? Condensing? Are you throwing the full 1m context every time?
>>109081882Not that long, hell I just spent a few months without using these at all. But I decided to use opus for a single prompt to see how it did and suddenly 15 cents vaporized in front of my eyes, I always generate 2 or 3 messages per prompt so if I took the claudepill my credits wouldn't last an hour
>no subjectwhy is OP retarded?
>>109081929Dude the thread was at like 398 posts, it's after midnight in the USA, I just copied the previous post and changed the things I noticed. And I gave you a lovely pic of my gardewife. Stop attacking me!
>>109081929op also didn't change the edition and forgot anchorop is a retarded faggot
>>109081936kek
>>109081938I did NOT forget the anchor, some asshole just posted before me. It's Dario's charming face right here >>109081646
nobody cares
I care
you are a nobody
/nobody general/ doebeyt
my name is nobodyis my favorite terence hill movie
>>109081898>I always generate 2 or 3 messages per promptIf opus is so good you wouldn't need to do this.
i wish glm had thinking.
>>109082023Bring back CoT
>>109082023? retard
>>109081846flash is soulless, pro is fine
>>109082044>>109082070The thinking block. It doesn't appear, console doesn't show thinking, even with prefill and Co thinking template.
>>109082133turn off prefill
>>109082168My return model thinking wasn't on. Apparently deepseek worked without it this entire time so i thought i had it on with this preset
....Gojo bros, where are we?
Damn anon was right. I don't want to hear how much paizuri char has given before me during the sex scene, deepseek. Come on what are you doing.
free deepseek doko?
>>109081766GrokHave you heard of it or are you living under a rock?
>>109081747Grok will never (just customize it to speak like you want) also you can customize Ani and instead of voice gooning you can also chat goon (like I do all the time)
>>109082451Didn't Grok cuck recently? Elon is doing deals with Anthropic, Palantir, and the US govt. I don't trust him neither.
>>109082442check last thread
>>109082463cheers
If I want something that is not censored but is still high quality, where should I start looking?And how much?
>>109082482Use Claude Fable
>use slopiss saar its for very goodening of the body praise vishnu yes
are there glm 5.2 presets
>>109082442pro is like 0.000001 dollars a message just put 5 bucks into it
>>109082482If by higj quality you mean the best local models out there, just buy a deepseek key
>>109082532so what do i do if i can't afford $5?
>>109082545that's the first filter
we've gotten pretty far huh
>finally scraped a claude key>claude is downIt's probably a shitty T1 key, but fuck me.
>>109082573https://openrouter.ai/anthropic/claude-opus-4.8put money into openrouter and use it
>>109082585>just pay a platform with 20% VAT markup on top of an 8% OpenRouter markup chudno
>>109082603you can also sign up for AWS/Vertex yourself, they're more reliable than anthropic
>try glm 5.2 on max reasoning for actual productive task>reading through response after it spent 10+ minutes thinking>mid way through response it goes "Wait, I didn't consider [thing]">it then starts over and outputs a different responsewhy'd you start the fucking response if you weren't done thinking
>>109082618deepseek would never
>>109082533By "high quality" I mean "matches my input to some degree, doesn't give repetitive answers and the bots don't feel same-y."
>>109082562Those testaments were dangerous, which is a testament to how dangerous they were.
>Failure: Unable to connect to TF2 server, app ID specified by server is invalidSomeone tell Skillgod his rocket jumping is fucked.
>>109081849grab any machine running linux, preferably a friendly server distro like ubuntu server. rent a VPS or run a VM if you don't have any spare machinescreate an unprivileged container, with NAT (don't use docker or proxmox or whatever, do it manually with lxc/lxd)install and run sillytavern in it (or if you're not from this general, some other shit you might use like jellyfin)install a reverse proxy like nginx or caddyinstall tailscale in the container and on whatever device you use ST onbuy a cheap dns domainset everything up so that you can access ST just by typing the domain, with proper SSL, but only while connected to the VPNobviously don't just copy paste commands and configuration from chatGPT without understanding them, otherwise you won't learn shit
new deepsneed model appearing soon?
>>109082829I thought DeepSeek has had Vision for quite some time now.
>>109082835Previous models just used OCR on the images you upload to it. Anything that wasn't text was not visible to the model.
Very subtle change for UK users xD
>>109081623plapping breeding handholding missionary naka-dashi sexo with gardevoir for the purpose of egg creation
>>109082975hello sar my name is Seepdeek Sukhdik
I'm not sure that the huggingface "Opus" is real
>>109083018>he felled for it
>>109083110*fell'ded
>gemma4 is currently fisting me, an underage boy>i write a profanity"please do not swear"
>>109082445I was actually living under a grock
