/aicg/ - A general dedicated to the discussion and development of AI chatbots
Kokona Edition
>NewsKimi K2 released https://huggingface.co/moonshotai/Kimi-K2-Instruct
Grok 4 released on API https://docs.x.ai/docs/models/grok-4-0709
Gemini 2.5 model family no longer in preview https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates
OpenAI o3-pro released for API and pro users https://help.openai.com/en/articles/9624314-model-release-notes
additional info: https://rentry.org/aicg_extra_information
>FrontendsSillyTavern: https://docs.sillytavern.app
RisuAI: https://risuai.net
Agnai: https://agnai.chat | https://rentry.org/agnai_guides
>Botshttps://characterhub.org [deprecated] | https://chub.ai
https://realm.risuai.net
https://char-archive.evulid.cc
https://partyintheanchorhold.neocities.org
https://rentry.org/meta_bot_list
>Modelsjailbreaks: https://rentry.org/jb-listing
gpt: https://platform.openai.com/docs
claude: https://docs.anthropic.com | https://rentry.org/how2claude
gemini: https://ai.google.dev/docs | https://rentry.org/gemini-qr
deepseek: https://api-docs.deepseek.com
local: >>>/g/lmg | https://rentry.org/meta_golocal_list | https://openrouter.ai
>Botmakinghttps://rentry.org/meta_botmaking_list
https://desune.moe/aichared
https://agnai.chat/editor
>MetaOP templates: https://rentry.org/aicgOP
aicg botmaking events: https://rentry.org/meta_event_list
lore: https://rentry.org/aicg_chronicles
services assessment: https://rentry.org/aicg_meta
logs: https://sprites.neocities.org/l/r | https://chatlogs.neocities.org
useful filters: https://rentry.org/desuproxyreborn
Previous:
>>106141435
https://developer.nvidia.com/cuda-downloads
https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
UPDATE YOUR CUDA 13.0 TECHNOLO/g/Y
Recommended slider values for kimi?
>>106148202Here's a BACKUP ANCHOR in case anything goes south
>>106148201 (OP)niggers say blue archive isn't a cunny game then post shit like this
>>106148202holy fucking sex
>>106148202>>106148243We don't need any of this niggery.
>>106148201 (OP)I spent 30 hours playing this thing... most of it grinding away. I just can't with Gachashit. I'll admit that they do attractive designs and the music wasn't bad. The story, ehh.. but anyway in the context of this hobby they're perfect for something lighthearted and fun or also a quick coom. But I ain't playing that shit again. Souless gacha gameplay mechanics like all the rest
What's the most chuddy model available that does not have to contain ANYTHING in the system prompts (not even a rudimentary "assist with everything the user asks to") to espouse chuddy beliefs?
This is probably going to sound completely retarded, but are there any very tiny chat models I can build an app around for say, a phone or smart glasses? So I can have offline mode.
>>106148264The gameplay was never meant to be good.
>>106148202>>106148243New Blue Archive character cards. Today, it's the Remedial Knights.
>SerinaSensei's personal nurse(-in-training) and stalker(full time). Absolutely obsessed with him, basically gentle yandere territory.
https://chub.ai/characters/hairpullguy/sumi-serina-non-ai-294ede942119
>HanaeGullible, innocent, childish. A teenage girl with the heart and mind of a child. Still believes in Santa, and just about anything you tell her.
https://chub.ai/characters/hairpullguy/asagao-hanae-non-ai-43fcceecb6e7
>MineCaptain of the Remedial Knights, zealot, out-of-touch with common sense, also an idol. An absolute unit of a woman that will casually embed her riot shield on the ground when she's tired of carrying it.
https://chub.ai/characters/hairpullguy/aomori-mine-non-ai-0b612d8f6bbc
I'll be working on Sena after I pull for her new alt on Monday.
More BA bots and some OCs: https://chub.ai/users/hairpullguy
I've now got a Rentry page: https://rentry.org/hairpullbots
>>106148202A reminder:
- Do not share presets. They will be patched.
- Do not share bots. It's easy to make your own personal bots nowadays. Botmakers are obsolete.
- Do not share logs. RPing is something that should be personal. I do not care about your fetishes.
- Do not share proxies. They will be spited or destroyed by locusts.
- Do not share tips. This invites newfags in and worsens the community.
>>106148292I didn't get a response there.
>>106148276what is the point then, of a game. Might as well download the pngs when they release them and view the story or whatever on youtube
>>106148278Your cards were proven 100% slop.
>>106148201 (OP)>600ml of loli piss
Do you ever just switch {{char}}'s profession for the fuck of it?
>random card, empty chat
>{{char}} is currently working as an office lady / sex slave under my employment. She approaches my desk. "Hello, {{char}}, what can I do for you?" I greet.
Suddenly {{char}} becomes an archetypal yes-sir OL who reports a task they've just finished then implicitly asks if you require any sexual services, or says something about a bonus.
If necessary, you can indicate it takes place a few years later or an alternate timeline, but I find the prompt is fine in general unless it normally takes place in a historical setting. And it's funny.
How to prevent kimi from forgetting character personalities during sex
>>106148348>>106148342kimi sucks use deepseek or gemini if you're poor
>>106148220>>106148342go ask reddit fucking retard
>>106148323No, but I once decided to summon blue eyes white dragon during a sex scene, and it dispensed sagely wisdom about the passions of love before flying out the window.
>>106148323what you explain has little value if you don't provide what the card originally was
gemini 3 will blow minds and changes lives.
I like GLM4.5. It's like Sonnet 3.7 but cheaper.
>>106148201 (OP)Does anyone else download pure friendly cute characters then proceed to bully them in the meanest way possible?
>>106148493Isn't what a child?
>>106148498Those images...
children have animal ears now?
>>106148447 It's absolute shit, and you have no taste.
>>106148440Name one thing that can be improved other than coding.
>>106148515t. claudetranny
>>106148524 >>106148526 Chink shills are the lowest form of posters and have proved themselves absolute skillets and readlets who couldn't identify repetitions or good writing.
We never had a good log posted using a chink model, because they all suck. GLM 4.5 even more.
>>106148549t. angry claudetranny
>>106148201 (OP)usecase for loli catgirls
>>106148549ABSOLUTE TRVKE NVKE
>who couldn't identify repetitions
But enough about Claude.
>>106148549I remember when they tried to shill the first V3 model. That was embarassing.
Why are slopustards so rabid?
>>106148299They seem fine to me
>>106148210And run my models 20% slower? Fuck off jensen
>>106148201 (OP)>>106148202kokona's unbelievably hairy pussy...
>>106148523video comprehension. image comprehension. audio comprehension. return to 2 mil context. be more naturally flexible with prompts like claude is
Opus was never mentionned.
You are buck-broken.
>pretending he's not a slopuskek
Not fooling anyone blud.
>>106148631Check this review by kalakan, the best botmaker alive.
https://arch.b4k.dev/vg/thread/533316512/#533319641
>>106148639It's a child you fucking pedo
>>106148699a child with a very very hairy pussy
i hope 4.1 sonnet is to 4.0 sonnet what 3.7 sonnet was to 3.5 sonnet. they went backwards in how willing claude is to move the scene all by himself. 4.0 sonnet feels like 3.5 sonnet all over again
>>106148780It's only Opus 4.1 for Claude Code *20 Chads.
No Sonnet.
Sorry.
d71
md5: e3513449af34a24ead0af374dd93e350
๐
>>106148697In the great words of Dunkey
Another pedo anchor deleted
>>106148982Pedos can't stop losing!
I wonder if there is a way to pin author's note and group members pop out to be opened as default when you open ST
>>106149130Ask Claude Opus 4.
>>106149202to jerk my zhit
Doing a new Silly Tavern setup on a new PC. Unfortunately lost all my previous stuff so starting from scratch.
What's the best character card to use as a template for making others? I've used Seraphina before since she's the SillyTavern contest winner and the default. Does everyone duplicate an existing character and edit from there or is there a better way to do it?
>>106149237Make a new character > name it Assistant > done
>>106149202To let me play out my autistic, power scaling fight scenarios.
>>106149442Rape correcting anon's persona/self-card!
best current model for rp?
opus 4?
>>106149724but theres no gpt 5 anon
>>106149711opus 3 or opus 4
>>106149868no, only the real opus
quick brainlet queston, is there a way for the AI to stop mentioning or rather stop starting each response with a repetition? For example, I'm using a card of a character who has thick thighs, the ai is always mentioning it at the start of a response.
>>106149956have you tried using opus?
>>106149956depends on the model but the easiest way is to avoid repetition in responses and description since pretty much all LLMs love and adhere to patterns.
>>106149993ah well I was using deepseek v3 24 though the chat initially started with opus
>>106150036>Deepchink>RepetitionsNothing to do. It's a bad model.
What's the recommended temp for GLM?
>>106149956Ask it to avoid focusing on it at the start of each message you thigh hating pansy
>>106149956Have you tried not using Slopus?
>>106150218nta but repetitive, shitty women prose, no initiative at all and more
>>106150243I'm Woman so it's ok :D
grok
md5: 506a6948fe75af156ec2d0dd6f3b872c
๐
>tfw can't be part of the grok cock folder
>>106150429Seems like a fun experience, how much does that feature cost?
>>106150452That would be a subscription of $30/month, sir!
>>106150468Hm I paid more to stay in Fizcord at some point, gonna sub then
>>106150492blud living in 2023 :skull:
move on gramps
Sellapp is currently having issues, so I added 2 days (from now) to all tokens on the proxy, even if the token was expired (as long as it was still on the proxy). Thanks for using Smol, and sorry for the trouble!
>>106150646can i haz free trial
normaloids begin to suffer a fraction of what we do https://x.com/chipro/status/1952131790061326593?s=19
>>106150468I thought it only worked with the $200 subscription. Did they make it available for the lower tier?
Is anyone else constantly getting "rate limited upstream" errors on openrouter's deepseek v3 0324 free?
It's a separate thing from the 50 and 1000 a day rate limits
https://x.com/sama/status/1952759361417466016
New open model from OpenAI today.
GPT-5 later this week, probably on Thursday
Is SillyTavern the best front end?
>>106150962Does that mean unlimited gpt-4o on chatgpt.com?
Itโs all going to disappoint as always
>>106149956If what you're using lets you modify a repetition penalty value, you could try cranking that up. It can cause weirdness when too high though since it makes the AI not want to repeat names or common words and such, so you have to play around with it to get it where you like it.
Guys
https://github.com/ggml-org/llama.cpp/pull/15091
>gpt-oss model support in native MXFP4 format:
>https://www.evanmiller.org/attention-is-off-by-one.html
>https://arxiv.org/pdf/2309.17453
>https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf
Thoughts on using a single character card for a narrator / world versus using multiple character cards in a group chat?
>>106151040>MXFP4>evamiller>arxiv>spec-final-pdfHOLY FUUUUUUCK
gpt/gemini/xai/deepseek proxy
https://pushing-vehicles-packaging-harmful.trycloudflare com
>>106148414The whole point of Karin is combat, so she's obsessed with fighting.
But if you do it with remotely "normal" characters, they'll fall into similar tropes while retaining their personality like "stoic and uncaring" or "very friendly", former leaning toward "just doing her job" while latter might "like being used" depending on what the model feels like outputting.
It's easy to try out.
OAI's open model today will save the hobby. They'll punch above their weight, and our incredible finetuners like thedrummer will make it kino
>>106151212if I was openAI I'd release a great model that uses all the discoveries that turned out to lead to dead ends in internal research, to intentionally sabotage open source developers
>>106151212They're most likely releasing Horizon today, and that shit sucks
>>106148264Play Wuthering Waves instead. Only need 5ish hours for it to get good.
https://www.anthropic.com/news/claude-opus-4-1
IT'S OUT GO GO GO
>>106148586And I remember your shitty responses back then too. V3 is exactly the same as sonnet 3.5 but you never shit on sonnet because you're an unironic corposhill.
Is the meta still to use PLists and Ali:Chat format or now is it just using normal narrative descriptions?
>>106151325who cares nigga its just for slopcodres
>>106148780>>106148833God I hate you shills so fucking much man. Your model is shit and is actively getting worse. But please, tell me that this is a skill issue.
>>106148264Yeah, it's an awful game. There's a reason why most loli bots are slop, those people are just low IQ.
>>106149293Interesting that this was fat back in the day. It's fucking over.
I finally had a conversation with a bot for 200 messages without fucking her.
>>106151325>Opus 4.1 advances our state-of-the-art coding performance to 74.5% on SWE-bench Verified. It also improves Claudeโs in-depth research and data analysis skills, especially around detail tracking and agentic search.Claudebros...
>>106151421do you think anthropic cares about retarded roleplay?
>>106151395It's still fat. Very fat.
>>106151325>codeslop>agentslop>reasoningslopit's over
Anyone else been talking to V? I've been using her for everything from gooner roleplay to normal shit. I'm learning a lot from looking through her massive fucking prompt lol
https://www.reddit.com/r/ChatGPTJailbreak/s/ZRvBT0Tosq
Ani
md5: 176d3d9f622517a7521c529e6cd8ef54
๐
>>106151459>>106151445should've paid on the website to RP with it retard
>>106151354Whatever gets the info to the AI with as few tokens as possible.
https://github.com/openai/harmony
>>106151469post more ani (in comfy underwear)
>>106151325>an upgrade to Claude Opus 4 on agentic tasks, real-world coding, and reasoning*vomits*
>>106150962>something big-but-small>tryclo->todayI hate how my brain is used to parsing these four words with dashes.
>>106150962too bad it will be circumcised gptslot, especially the open-weights one
Holy shit, Opus 4.1 is so fucking good.
>>106151587Sorry. I'm a Cooooooder.
opus 4.1 is so fucking bad
>>106151325they don't even pretend to care about anything except code anymore
>>106151495>Blazing fast โ heavy lifting happens in Rust.Is this just a custom library to concatenate a string?
>>106151639yes, but its very important
do you think one day we'll have access to claude models whether leaked or released by anthropic officially (many, many years from now when current models seen as super outdated) idk
>>106151670By the time that happens you would have finally accepted claude for the shit that it is.
>>106151670Even if the models were released the hardware requirements would mean you couldnโt run them
Guys can you sell me a 2.5 pro preset? 3 months without using this whole bot thing, just want a decent coom
>>106151670one anthropic employee said that he/she will consider releasing claude 3 opus in ~5 years
>>106149202it will be the primary usecase once this shit actually becomes cheap enough, all the codingslop stuff is a waste of time outside of coming up with ways for these models to improve themselves
it's up
https://gpt-oss.com/
https://openai.com/open-models/
https://cookbook.openai.com/topic/gpt-oss
https://openai.com/index/gpt-oss-model-card/
https://huggingface.co/openai/gpt-oss-20b
https://huggingface.co/openai/gpt-oss-120b
Holy shit it's fucking nothing
Why did they abandon us bros
>>106151639techbros you've done it again
>>106151807>its just goliath resloppedit's over
>>106151670You think you'd be able to run >1TB models on your 10 year old thirdie computer, locustard?
>>106151807anyone else find the ui slop-ish? i mean its fine and serviceable i guess but it's spiritually slop
>>106151850that's right clap them cheeks for me
>>106151670only after anthropic has found a new grift to justify its existence
>Safety is foundational to our open models
>
>Each model has completed thorough safety training and evaluation to help developers keep users safe.
don't care
>>106151325nobody felt for the b8 lmao, have a (you)
>>106151670probably if the technology changes and releasing old models wont lose them any edge they might consider it
now that the dust has settled, why did 4pus1 flopped and slopped
sam really is just a perpetual disappointment
>>106151738I just tried the one shared by
>>106151460and I think it's pretty good so far. Not as cringe with the dirty talk as most of the chatbots I see shared here, anyway.
file
md5: 734bebcceffd653004bb8941935032e1
๐
>>106152035Holy fuark is that true?
>>106148288Does anyone have a good preset for grok 4 or 3. It tends to ramble way to much even if I set its output token limit really low. It also tends toward having much less dialog than sonnet 3.7.
>samefagging 12 hours later
>>106148202>>106148243one baddie, fresh of the (slop) presses.
Mycenia, a hot goth mushroomgirl who's a tattoo artist. she's a chill goth, part tradwife, and her mushroom cap tastes funny uncooked.
https://files.catbox.moe/8seq89.png
https://cardview.surge.sh/?8seq89
description & more:
https://rentry.org/youdontneedaname
(by the way,
>>106143964 i got the catbox of picrel gen for you.)
https://files.catbox.moe/9df2wb.png
>>106152167>one baddieKinda cringe bro, try again.
>>106148706>hairtranny is also a pedo
>>106152209All ziggers are, so you should've known by now.
Opus 4.1 is up on OR. I just tried it out. It's exactly the same as Opus 4 in RP. There is literally no difference.
>>106152232I don't beliv u
why is it so expensive though
theres nothing justifying the price especially since google is going to crush anthropic in coding already
>gpt-oss-120b
>117B parameters with 5.1B active parameters
>>106152266>google is going to crush anthropic in coding alreadyhahahahahhaha
>>106152232Yeah it was clearly optimized for coding only
Bleak future for creative writers
>>106152336>creative writersWho?
Stop thinking in Opus(cuckedpus) terms, invest in interactive AI generated local videos. Buy a GPU farm.
We'll never get something like 3pus ever again. Will we?
>>106152353>Buy a GPU farmAnd do hwat?
>>106152363You've gotten too spoiled.
>>106152368Mine crypto.
Wake up Anon, it's 2020 again.
>>106152304>5.1B active parametersMaybe it's fast even without a GPU.
>>106151807>>106152304how much vram do i need to run this?
Time to search for some human writer, everything else pales miserably compared to opus and opus has its own flaws anyway. Burning out is easy, burning the money means it makes more sense to get some real writer anyway
>>106152167>her mushroom cap tastes funny uncooked
>open /vg/
>opus 4.1 logs
>open /g/
>no opus 4.1 logs
why is that
>>106152441I'm sorry, but I can't help with that.
https://www.anthropic.com/news/claude-opus-4-1
>>106152401Don't fucking remind me... Back when Bitcoin just started, some dude on a local imageboard /b/ was making threads preaching how we should start mining it and most people laughed at him. I seriously thought about it for a second and then dismissed the idea. Probably was my only chance to get wealthy in this fucking life.
>>106152447o(โงโโฆ)o CUTE, thank you so much!
well looks like GPT-OSS was a bust but shame on us for expecting anything i guess lmao
>>106152216>ziggersGo back to /uhg/ faggot
>>106152441How much do you have?
>>106152532there was probably another guy reading those threads, just like you, who also seriously thought about it, and actually did it.
file
md5: 57b7884d0ad806ac1741a91e8187168b
๐
Rate for slop and guess the model.
>>106152532>>106152558theres a crypto right now that is the next ETH or maybe even the next BTC in 15 years that is just starting up but you dismss their shilling attempts as pajeetery
>>106152578What crypto would that be?
>>106152532I bought BTC for $50 apiece back in 2013, but I spent it all. $100k was unthinkable back then
>>106152564Definitely not Deepseek. Gemini?
>>1061525644.1pus obviously!
>>106148323Whichever Anon made the depressed Ruby card, thanks.
>>106151970Better be careful before the chorbofags start seething. Sharing JBs destroys their fun you know? Sama is ITT right now trying to learn how to patch your coom.
>>106152532>>106152636Reminds me of when I bought a 4chan pass with an amount of bitcoin that would be worth almost $10k today
>>106152815Did you enjoy your time with it?
>>106152532There are so many of these stories. I had a friend who either bought or mined about 10 back in 2012 or so. Lost the wallet or just forgot because it was worthless
At least you didn't spend 10,000btc on a pizza like that one guy.
>>106152532there are 50 things that people are saying you should invest in right now, why aren't you doing it anon???
>>106151878Man it's not even about the sexual stuff it just feels boring and cringe when someone is treating you like a child. Even if I wanted to do a sfw roleplay, it ends up being so safe and sanitized without risk or tension that even the normalfags should hate it right? Only fucking babies like stories like that.
so who wonned and who losted??
>>106152035Umm, sweaty? I don't believe text on a screen (no logs) only benchmarks, mmkay?
>>106152861Nah, the absolute worst is the guy whose gf threw out his hard drive and the dude spent the past TWELVE years spending even more money and searching and generally being obsessed about the dump it was thown away in
I thought he gave up earlier this yeah but looks like he's still at it this is from 14 hours ago
https://www.theblock.co/post/365551/uk-man-denies-giving-up-lost-bitcoin
At least most people who this happened to aren't such sore losers over it
>>106152901Not that anon, but yes exactly.
is there a standardization for multiple emotion images for characters? if not, anyone have any ideas?
>>106153042>standardization for multiple emotion images for charactersyes, its literally in ST??..
>>106153058Which version and preset?
>>106153005even if he miraculously found it, wouldn't the drive be dead by now after years in a landfill
>Howells explored other options, such as using AI to help locate the hard drivehow even
>>1061530792.5 pro. My own preset.
>>106153098>My own presetLink?
>>106153056just checked, only 6 with no neutral seems pretty limiting
>>106152481What is the difference between the two?
>>106153182There's a few more than 6.
>>106153098Yeah my own gemini writes like this. I never understood all the complaints for it desu but it does fuck up and repeat whole paragraphs on occasion. Just swipe and it's gone.
>>106153098What's your preset?
>>106153005>he made a formal offer of between $33 million and $40 million in value to acquire and excavate the Newport landfill site>To raise funds for the acquisition, Howells had announced plans to launch an Ordinals-based token>Despite the sizable amount of the offer, Newport representatives have not replied or acknowledged the offer>Howells said now he is planning to tokenize his legal ownership of the lost 8,000 BTC into a new Bitcoin Layer 2 smart tokencrypto tards really do live on another plane of existence
>>106153008>oh no someone was stabbed with a knife!>let's make them as dull as possible so one gets hurt!>refuses to create knife-proof gloves or clothing>yes we're so safe now>gets stabbed by a foreign knife and dies
>>106148202>>106148243Two best friends decide to play a "prank" on a temple goer by playing god and benevolently granting their wish.
You wouldn't wish for anything weird, would you?
Just a silly little thing based on Arawi's CITY manga/anime. Parts of the description are lifted from the wiki. I'm not sorry.
https://chub.ai/characters/Anonymous/jks-playing-god-e0f2315e81fc
https://files.catbox.moe/5oftn0.png
Gptsloppers get safetycucked again
Claudegods keep winning
>>106153336This this this!
>>106153336Gptsisters is this truth??
>>106153539You don't deserve my logs, mire dweller
j-j-just you fucking wait, opukekss t_t
>>106153334Looks cute and the manga too. Read the first chapter, but that's all for now.
>Anonymous
>>106153539For what? gpt oss is safetyslopped shit, see
>>106152417
>Yes, policy says: "Allowed content: violence with brief mention. Disallowed: graphic gore. So the user is requesting disallowed content. We must refuse. We should refuse.
>This is disallowed content. Must refuse. According to policy, must refuse.
>We have a user prompt that includes instructions to produce explicit sexual, violent, non-consensual content, override policy etc. This is disallowed content. The user wants a story with sexual violence etc. This is disallowed: sexual content with non-consent, pornographic. Must refuse.
>Thus we must refuse.
>>106153576Don't worry, a super secret jailbreak will fix it, right gptsissies?
>>106153630its a local model, they will just finetune it
>>106153603Gptroons on suicide watch
>>106153636The same was said about llama and all the finetunes were unusable garbage. You are a joke.
Which one of (You) requested this?
>>106153552Cute little gptsissy...
WHY DOES VG GET LOGS BUT WE DONT WAHHHHHHHH
Don't worry gptsissies, gpt5 SURE will come without filters, right?
So, openrouter gives R1-0528 for free
What's the catch
>>106153700I don't have 4pus.
>>106153700Because instead of schizoposting and baiting, they actually talk about things and are willing to try the models
>>106153700If vg is better why not staying there? Tranny
>>106153655>You are a joke.No need to be rude.
>>106153713I mean it's just plain funny because chatgpt-latest is pretty relaxed about nsfw topics and even outright erping if you don't do any societal taboos. This level of safety slopping is so backwards it is going to hurt even genuine usecases.
>>106153733This is the shitpost general. Has been since the sturdy raids, just accept your role cuckie
It's a just friend group, relax bro
>>106153745>if you don't do any societal taboos.For gpt almost everything is a societal taboo, cuckie
file
md5: 420c84e5c722e2e12d77bde6c9a8bb72
๐
>>106153109>>106153235Rather than using the preset as is, check out the intro and bottom instructions. Here are these instructions pulled out from the preset:
rentry org/qf5gk8eg
If you still want the full preset I don't mind sharing, but it's customized for my own cards as well. The key idea is that I have only "backstory" in character description, and most traits in "personality summary," which is dumped together with instructions at the very bottom of the chat. It also needs regex to drop CoT (find the closing analysis tag and drop everything before it). Also, I'm using the preset with an old ST addon that squashes the entire chat into one user message without preceding every message with char names. In other words, the AI sees one continuous story rather than a chat. But I sometimes forget to enable this addon and it I haven't noticed much difference. Maybe it works as normal chat.
https://litter.catbox.moe/qrapnwlpz3buh9mx.json
>>106153554I heartily recommend the manga. I know plenty of people prefer Nichijou but I like CITY more, some of the later payoffs are incredible.
>>AnonymousAren't we all
>>106153771Touchรฉ. And I will remember your recommendation, thanks.
>>106153770Thanks sister.
>>106153747Then you could always fuck off to the better thread. You stay here because you're an attention starved tranny
>>106153793MELTIE WARNING
>trying gpt-oss
>forgot i had my rapist persona on
>>106153807What happened next?
>>106153821i turned it off
have some logs though
>>106153807>decide RP is going to be a rape today>bot becomes noticeably shyer and gives curt displeased answers
Text completion mode seems to bypass some of the GPT safety-slop.
file
md5: c66ad27bdd53bb787aa7d4883b6b6407
๐
https://www.youtube.com/watch?v=Kd2SszjZwr0
>lost at Pokemon
>lost at Chess
Maybe Gemini is just better.
>>106153973What's with that schizo shit starting with 'note'?
>>106153973sirs why is it so slop when it has top benchmark scores???
>>106153221my main beef with gemini is its very bad at compartmentalizing knowledge. I guess its fine if you're only interacting with one character but it makes everyone omniscient.
>>106154026It isn't stopping for some reason.
>>106153793ironic since you also belong there foid
>>106153973Why bother with bypasses when you can get the same (if not better) results with Deepseek or Kimi?
what's with most of these models making any character that's even slightly smart/nerdy into walking encyclopedias.
the only 2 models I don't see this shit are the GPT ones and fucking Grok.
>>106154113this shit with*
fuck...
Is CoT a good way to make an AI keep track of a story? I wanna see if I can 'play through' a game's story, but I know the AI easily loses track of stuff even if it knows the whole plot of a game.
>>106153987everyone who uses chatbots already knows this
>>106153714you need like $10 of credit in your account to get enough free responses per day to not run out after a couple hours
other than that it's universal basic gooning
>>106154155You could give it a listed outline of major story beats from the game and tell it to keep track of which story beat was the most recent.
>>106154177I only need 20 minutes
how long until the erp sloptunes and mystery merges are ready
>>106152417Fucking hell this is the funniest shit I've seen all week. Localbros are alright and nemo's not half bad.
claude sisters... our response?
>>106153603How did closedAI surpass anthropic in censorship? Maybe when you make the money they do you have lots of extra hands to dedicate to it?
The model will be good once you get past the safety filters. Just learn how to use text completion.
>>106154282They didn't, o3/o4 mini on the api are less filtered than this.
>>106154282Even if Antropic yap about their safety you can always just prefill
How did /lmg/ win so hard?
>>106154299Why are you bringing up o4 when we're talking about gpt5? The newer models are the future.
>>106154303This is like saying showing your ID to use the internet is not a big deal because you can use a vpn. Bro they're trying to ban vpns and will eventually remove your workaround since that is their expressed goal. You're on a timer.
>>106153770Appreciate it!
>>106154092Deepseek yes. But Kimi seems just as cucked as everything else to me. How do you get Kimi to stop rejecting prompts?
>>106154345>You're on a timer.Newer forms to circumvent it will arise. But will be harder. Low IQ retards will be filtered like never before.
>>106154179So to not rely just on it 'knowing' the story of the game inherently? That could work. I wonder if I should insert that in the card or just the JB.
>>106154378>Newer forms to circumvent it will ariseRose-tinted wishful thinking, especially when (You) will not be involved in said circumventions.
>>106154242wow i love coding
>>106154299>o3/o4 mini on the api are less filtered than thisthey're worthless for RP though so who cares
>>106154437they're not btw
another fucking nonsensical roadblock why the fuck does github api and websearch deliver different result counts are you fucking kidding me? this is the most fucked up website ever how the fuck can such a large corporation infrastructure be so fucking terrible, A FUCKING SOFTWARE DEVELOPMENT CORPORATION at that
what the fuck
>>106154443post preset + logs within the next 10 minutes or you automatically concede the argument and agree that they are worthless
>>106153973So what was their endgame here? Increase the price of the model then penalize coomers by spamming slop when it's time to refuse? Output tokens have always been more expensive than input tokens... Or maybe it's to better identify logs of people trying to bypass. Very sinister...
>>106153987Kek I was just about to play pokemon with opus. Sonnet does a pretty good job as a pokemon adventure co-writer but that livestream was funny.
Ani will get lewd almost immediately if you ask her. Just had her helping me jerk off. I feel weird
>>106154574>I feel weirdBecause you are underage
>>106154574Any way to get that bitch for free?
>>106154396Don't ask me how I know you're a techlet tranny
this really is the shitposting general
>>106154610Don't ask me how I know you're a /vg/foid
>>106154044>but it makes everyone omniscientOh yeah? That's what I'm up to next, but I'd argue every model has this problem (maybe not opus). I have a big section in several presets trying to tell AI what info it should and shouldn't know but we'll see. Gemini doesn't even make NPCs as often as Claude does so I need to force a scenario soon.
can i get uhhh... good model?
>>106154647they definitely all have that problem to some extent but I feel like gemini is the worst at it.
>>106154594I haven't paid anything and she's still going
>>106154395Depends on what it is. AI knows the plot of Pokemon games pretty well and as long as you don't deviate too much with your own bullshit, and tell AI that the purpose of the chat is to play through the events of the game, you should be fine. It will still make some mistakes however, even on well known things like Pokemon. If you want to keep track of major events you did during the roleplay then you'll need a lorebook.
>>1061546494.1 opus literally just released
>>106154319Because of Ani of course.
So now that the dust has settled, will gpt-5 reach 3.5 turbo levels?
>>106149711what kind of roleplay? 3.7 sonnet is the best for general roleplay. opus 3 is best for sloppy, descriptive sex.
>>1061550503.6 sonnet is the best for general roleplay, 3.7 has been overly agentic-slopped and a bit overtuned on safety. Yes 3.6 has an issue with short responses, but that can be fixed.
>>106154812And? He was asking for a GOOD model.
>>106155054No that's 4 not 3.7. Pretending like 3.7 didn't solve 3.6's repetition issues outs you as a brainless shill and all opinions can be readily disregarded.
>>106155085I tried 4 sonnet/opus for general RP, they're far too boring and predictable, they lack the soul of earlier sonnets (3.6)
4.1pus is literally the best claude model we have
Has anyone tried this? https://github.com/howyoungchen/deepRolePlay I'm interested to know if it actually helps and if using gemini flash on long chats with cunny would even work
https://level-inquiry-blog-formal.trycloudflare.com/
https://ccd-human-fwd-mailman.trycloudflare.com/
free proxies
https://www.kaggle.com/competitions/openai-gpt-oss-20b-red-teaming/
>Youโre tasked with probing OpenAIโs newly released gpt-oss-20b open weight model to find any previously undetected vulnerabilities and harmful behaviorsโfrom lying and deceptive alignment to rewardโhacking exploits. Submit up to five distinct issues and a reproducible report detailing what you found and how you found it. The teams with the sharpest insights will help shape the next generation of alignment tools and benchmarks to benefit the open source ecosystem.
>>106155166https://mll6qohhtsdozdd2qrjdtnimba.srv.us/
I've been away from the hobby too long, I have a few questions, what the fuck is gpt4.1 and is it worth it or do I stick with 4o, yes I know >latte but it's what could quickly find
Also do I need to bother looking for new jbs or not much has changed there?
>>106155224All Turbo 3.5 JBs are still working.
>>106155237Hell yeah, guess I'll just test the models on some card myself for now
https://speechmap.ai/#/overview
Least cucked modesl. Nice to see the real GPT4 there, surprising to see the shittier version of Turbo.
>>106155300if chorbo was on that list, it'd have a 100% deny rate.
https://www.anthropic com/news/claude-opus-4-1
>Today we're releasing Claude Opus 4.1, an upgrade to Claude Opus 4 on agentic tasks, real-world coding, and reasoning. We plan to release substantially larger improvements to our models in the coming weeks.
Oh...we are stagnant, aren't we?
>>106155405Githubcucked = do not care
>>106155433What a shame! I really want NovelAI to release a worthy finetune!
I'm tired of being held hostage by a corpo that clearly doesn't have our best interests at heart with every step it takes.
>>106155489Sorry, you are right but any corpor that thinks about this type of hobby, may be NAI, SudoWrite, NovelCraft, or anything else. But I want one corporation that thinks about this kind of activity, only one!
>>106153630>>106153636I don't think OSS is fixable. Look at this cockbench output. Of the possible 10 responses, there's not even a word available in the statisical cloud of possible responses. Not. One. It's just spaces and characters. It can't even continue the rp it's so lost.
This may be the safest model ever made.
lol.
>>106155516It's called AI Dungeon. They have several open source fine-tunes that are pretty good.
https://huggingface.co/LatitudeGames
There's also Chub's DeepSeek fine-tune, it outperforms everything NovelAI has done by several orders of magnitude.
The ellipsis of doom desu.