Welcome to Chatbot AI General #100, the thread for discussing and improving AI pony chatbots.▶ MLP Botshttps://mlpchag.neocities.orgSpreadsheet (CAI bots + Old repository):https://docs.google.com/spreadsheets/d/1J7BeqJVDS51cXF8Pgm2YZaFq-Z6ykSJTCAI bots converted to Tavern: https://files.catbox.moe/ckurq1.zipExpression packs: https://rentry.org/ChagExpressions!!!GALLERY!!!: https://drive.google.com/drive/u/2/folders/1Ao-h5HFGMPllSrzSBKM_BvGSiU9f0c2U▶ How do I start?1) Select a Frontend2) Select an AI model3) Select Jailbreak4) Select bots5) Lovemaking with AI mares!Starting in this hobby can be confusing and difficult. If it’s your first time and you’re lost,▶ ASK THE THREAD! ◀Novice-to-advanced guide: https://rentry.org/onrms▶ SillyTavern (preferred frontend)https://github.com/SillyTavern/SillyTavernOn Android: https://rentry.org/STAI-TermuxApp that voices pony responses in ST: https://drive.google.com/drive/folders/16Ss26VBmgzcSuTGzhaHqRuyVRceTf-YB▶ More frontends:Risu: https://risuai.xyzAgnai: https://agnai.chat▶ Localshttps://rentry.org/lunarmodelexperiments>>>/g/lmg/Mistral Nemo base model fine-tuned on fimfics: https://huggingface.co/Ada321/Nemo_Pony_2/tree/main▶ JailbreaksMLP JB: https://rentry.org/znon7vxeMore JB and guides: https://rentry.org/jb-listingHypebots for Tavern: https://rentry.org/pn3hb▶ BotmakingEditors: https://agnai.chat/editorGuides: https://rentry.org/meta_botmaking_listAdvanced: https://rentry.org/AdvancedCardWritingTricks▶ /chag/ dustbin of historyhttps://rentry.org/ChagArchivePrevious thread: >>41638289▶ Current theme: Neet/MissfitNews:- Pixtral large out https://mistral.ai/news/pixtral-large/- Gemini-exp-1114 is now available via the API. Free on Openrouter too.- Anthropic retire Claude v1 models https://docs.anthropic.com/en/docs/resources/model-deprecations- 3.5 Haiku out https://www.anthropic.com/claude/haiku
Anchor for bots, lorebooks, scenarios.Previous >>41638292
Anchor for technical stuffs (Proxies, Updates, Models etc.)
Anchor for asking for bots, lorebooks, scenarios etc
Bots corner:Neet Twilight Sparkle >>41641265 https://mlpchag.neocities.org/view?card=Anonymous/Twilight%20Sparkle%20Neet.pngZabani >>41642592 https://mlpchag.neocities.org/view?card=littlewitcher/Zabani.pngZephyr Breeze >>41644461 https://mlpchag.neocities.org/view?card=Snowfilly/Zephyr%20Breeze.pngArt corner:>>41649095 Screenshots corner:Partial Alicornification Woes >>41638893Pantsed >>41639299Unhinged >>41640240Neet Twilight Sparkle >>41640844Tactical >>41645828
Screenshot corner 2: Daddy >>41647094[Activity you mentioned] >>41647768Rarity loves you >>41650284New Gemini Test >>41650732Horsepussy needs >>41656475Hatefucking Rarity >>41657080CYOA preset >>41657480
Anchor for the Neet/Misfit event!Everything goes as long as it’s on theme.Post bots, post logs, post lorebooks...End of event: 11/27>Neet Twilight Sparklehttps://mlpchag.neocities.org/view?card=Anonymous/Twilight%20Sparkle%20Neet.png>Zephyr Breezehttps://mlpchag.neocities.org/view?card=Snowfilly/Zephyr%20Breeze.png
>>41658078Reposting rentry.org/CharacterProvider-CYOARPGFun preset.
>>41658114can confirm it works real nice combined with other cards like Pony Colony Simulator
LOVEMAKING WITH AI MARES!
>>41658161Based * 100
I believe in mysterymare
Happy 100th thread /CHAG/. Have some AI mares.An update of the Candance card I released for Valentines Day, with rewritten definitions and greetings, plus some new ones. Cadance: https://files.catbox.moe/z13v5f.png- Updated greeting: A tired overworked Cadance greets you in her throne room.- Updated greeting: Cadance breaks into your house (again) to review the results of her matchmaking survey.- Updated greeting: Cadance finds you all alone on Hearts and Hooves Day.- New greeting: Comrade Cadance seizes the means of affection.- New greeting: Casual chit chat with the pretty pink pony princess at a pizzeria.New mares:Sassy Saddles: https://files.catbox.moe/otg5zv.png- Sassy greets the newest customer at the Canterlot Carousel.- You’re Sassy’s new assistant.Tree Hugger: https://files.catbox.moe/tlm65m.png- Hanging out at Tree Hugger’s cottage.- Hiking with Tree HuggerSongbird Serenade: https://files.catbox.moe/wtkmmg.png- Songbird invites a lucky guest backstage for a meet and greet.- You run into Songbird Serenade at an after party.
Happy 100th fellow anons. May the lovemaking with AI mares continue ever onward
>>41658218Holy fuck.Respect anon.
Are these settings okay for Claude 3.5 on Risu?
>>41658300Maybe disable everything but temp?
>>41658300Temp and Top P are fine. Top K at zero works well with those settings. Claude doesn’t have penalties, so everything else shouldn’t matter. It’ll mostly be what’s in your preset that will have an impact now.
>>41658300Claude supports only temp, top_p and top_kGPT supports temp, top_p, frequency_penalty, presence_penalty, seedGemini supports temp, top_p, top_k -> and according to docs frequency_penalty, presence_penalty, seed but afaik current models do not support them. maybe Gemini 2 willfor the actual settings. the common agreement upon Claude is:temp: [0.9 - 1.0]top_k: [0] (disabled)top_p: [0.8 - 1.0]lower temp and top_p for more predictable/less-random generation. generally you most likely will be using 1/0/1
>>41658300Penalties do nothing to Claude. But everything else looks okay.
https://arxiv.org/html/2411.05000
>>41658347Why wasn't C3Opus considered?
>>41658347llama bros...
>>41658353 Good question. Maybe it was too expensive for the researchers, kek. I mostly posted it to show an approximation of how much recall can degrade with context depending on the model. Also worth noting it’ll be worse for RP since the events it needs to recall are vaguer than a direct Haystack approach.
>>41658218Hey! That's the Cadance who really wanted one Anon to be gay.
MysteryMare just went down
>>41658075>>41658218Forgot to anchor.>>41658369Now with twenty percent less gay accusations!
>>41658433I think it's was just my internet or something very false alarm
>>41658574nah it did go down for a little bit.
Claude is schizophrenic
>>41658626Not bad for Groq
>>41658138Tried this but the rolls seem to be borked and not properly using ability scores any help?
column's fucked again as usual.
>>41658347meh, needle in haystack (and its variations) is a very flawed metric, and they also test associative retrieval, not enumeration and not reasoning, so it's not going to help us gauge the real performancegemini visibly outperforms any other top model at retrieval (try RPing with it, it remembers everything), although reasoning over long contexts is what we actually want and every model is more or less the same in that
>>41658765Column have obus?
>>41658765I hope it gets unfucked soon
>>41658859Technically but on both sources it's borked, 3.5 Sonnet latest ain't bad though
>>41658864Wha columns link?
>>41658899
>>41658908It's surprisingly easy to find out the token list of a proxy once you identify their vps
>>41658666 The rolls are weird for me too. Sometimes it uses the D&D system correctly, and other times, like here, it adds the attribute score directly. But apparently, the author of the preset knows about it.
Tried running local models, any reasonable models refuse to work with 8gb VRAM and 32gb RAM. I'm probably doing some shit wrong, but I tried for hours to get it working.Proxies are getting shot left and right or are absurdly hard to get into.Is it over bros? Is it time to take out my wallet?
>>41659084Probably. afaik you can't really even get anything 'okay' running local without 16GB VRAM, and to get something properly 'good' you'd need even more. An upgrade or sub is probably the options if you don't want to go chasing proxies.
>>41659084 New Gemini seems to be free on OpenRouter: https://openrouter.ai/google/gemini-exp-1114 Hermes 405B is still free here too. >Proxies are getting shot left and right or are absurdly hard to get into. You can search on desuarchive for the domain name most are using and find a public proxy with GPT that way. >Is it time to take out my wallet? If you want Claude, try putting $10 on OpenRouter. Set the context at 15-20K and monitor your daily usage to see if it’s sustainable for you.
>>41659174Kek.
>>41658218All added and updated, thanks anon.https://mlpchag.neocities.org/view?card=Anonymous/Cadance.pnghttps://mlpchag.neocities.org/view?card=Anonymous/Songbird%20Serenade.pnghttps://mlpchag.neocities.org/view?card=Anonymous/Sassy%20Saddles.pnghttps://mlpchag.neocities.org/view?card=Anonymous/Tree%20Hugger.png
>>41659174Thought you could sneak a non-pony card couldn't you?
>>41656351Why wasn't this included in the screenshot corner?
Is there a Whoa Nelly card? I want to turn her into a snowbunny for Jaquavius-Anon
>>41659174Which preset?
For those using OpenRouter: ST implemented prompt caching for Claude, and someone made a guide: https://rentry.org/pay-the-piper-less If it works well, it should save you money. https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching
NOw that some time has passed, how's Gemini compared to Chorbo/Latte?
>>41659415 Less censored for sure (unless you do "uooohhhs T_T"? Not my jam, so I can’t say), but dumber, though less dumb than previous Gemini models. It’s just noticeable if you’re used to models like Sonnet 3.5 or Latte. You can have a preset focused only on writing instructions and still do NSFW from my tests. Writing style and prose can be good and fun, but it does suffer from older GPTisms from time to time (mixture of, testaments, etc.), which is really unfortunate. The base also seems less positivity-pilled and more proactive? Even if it’s still an assistant LLM, so it can suffer from that. I still prefer Chorbo, but if you’re tired of it or for not-too-complex cards and scenarios, I could see it being a good alternative.
>>41659385
>>41658071yay! happy birthday /chag/ and all you kinky horsefuckers!>>41659415agree with everything Anon above said (if you need an extra input)>>41659263finally, was waiting for this addition for quite some time. thank you for report, Anon>>41658218>Tree Huggerah, NEET pony, thank you for participating in themejoke aside, thank you for all those cards
when kirin lorebook
I see more JB's for Claude 3/2, is it more preferred then GPT 4 or o1? I might try it out if it is.
>>41659739 Claude was always the preferred model in /chag/, and it was widely available on public proxies before. Now it’s not as available, so many are going back to GPT, mainly the chatgpt-4o-latest version since it has the best prose. The JB list in the OP hasn’t been updated in a while, so it’s also lacking presets for 3.5, for instance. >o1 Not much used for RP because the "gimmick" doesn’t add much to it. For chatgpt-4o-latest JB, also called Chorbo or Latte, you have this, which was made here and works well for SFW and light NSFW: https://files.catbox.moe/yvlncj.json And this rentry for heavier NSFW: https://rentry.org/lattejb If you want JB for Claude 3.5, ask here. If you use OpenRouter, there may be some small changes needed.
>>41659764thanks man, i was just kinda being a slave to this table https://rentry.org/aicg_meta and I saw that Claude had some good ratings.
>>41659684do we have enough Kirins to even justify a lorebook for them? unsure about OC and fanon lore. as for their "quirks" you can just copy-paste description from mlp wiki, and post it in context if you think model doesn't get kirins right>>41659739GPT is an obedient dog which requires discipline, training and knowing how to handle itClaude is a cat who just does what it wants and it makes you wonder what is going on in his brainthey are two different breeds. for creative needs anons like Claude more because it can be totally schizo out of box comparing to GPT. GPT needs more time to get used to it and learning what and what it can do (and how)in addition to those two, we also have Google Gemini which is a different thing all together. more loose, less disciplined, more prone to drive story further, less smart. it is like a compromise between GPT and Claude
>>41660104I don't know why, but I'm getting the strangest feeling this test might not be entirely legitimate.
>>41660104Did you have to add stats to Cadance or did it just pull this out itself?
>>41660120 Using >>41658114, I just have stats for my persona, Anon. For the survey, it pulled it itself, there are instructions in the preset to display info like that when relevant. Also, it’s strange, but in edit mode, the questions are correctly numbered, but they all appear as "1." outside of it. Dunno if it’s a problem with ST or the preset/CSS. I don’t have any regexes that would cause that.
>>41660138Yeah the preset is alot of fun. Try giving yourself traits too. Only thing I don't like so far is the css not having chat bezel(?) out with the icon.
>>41660104Last one to not spam the thread, but it's a fun preset.
>>41660138>Dunno if it’s a problem with ST or the preset/CSSIt's on STT. I've run into it when cooding and planning out with some nested lists.
>>4165886017 hours straight and we're still on worst source. I miss claude.
>>41660474 It was on Source 1 for a while. >>41660138 The preset is really fun, but fuck, 3.6 is so lacking in swipe variety. I often get the exact same situations and CYOA options.
>>41658765I haven't been able to get column to respond at all. for a while, is it just roll till it works?
>>41660911On Source 2 is really bad, not sure if it works at all to be perfectly honest, I haven't ever gotten anything but a blank response on Opus or Sonnet on it at least. GPT works fine though
>>41660925Damn. Alright, thanks.
New mistral is ok
How do yoh talk with them?Is it as easy as ChatGPT via URL.
>>41661122use this to set them up https://rentry.org/onrms>Is it as easy as ChatGPT via URL.wellll not the url party (sorta) but the chat system works about the same, good luck anon!
Can someone help me?With the CYOARPG jb the text comes out like this, I don't know what's wrong, I followed all the steps of his rentry
>>41661208User settings, check off "show <tags> in reponse"
>>41661212Now I feel pretty stupidThanks anon
>MM is fucked>Agnai is fucked (Can't type any shit)Fuck it all
>>41661471>Agnai is fucked (Can't type any shit)Huh? It's still working just fine for me, so I think there's an issue on your end.
>>41661490Can't type username on login. Both PC and online.
>>41661471>MM is fuckedwha habben?
>>41661496Oh. Welp, guess agnaidev fucked something up when he pushed that imagegen update a couple minutes ago. But since it's something as essential as that, people will start to scream at him and thus he will fix it soon enough. Probably.
>>41661521It's not only at the login. Can't write shit on the library lorebook too.
>>41661561On one hand, the login page is working again.On the other hand, I can't see the lorebook being fixed without someone taking one of the team and submitting a bug report.
>>41661612>taking one of the team
>>41661616...in my defense, it's not easy to proofread when you are running on less than 3 hours of sleep.
>>41660180>LOLbased Celestia. Great preset anon, made me laugh. Will have to try it.
>>41661633Hello and good luck, myself!
>>41658908>>41658998what's column's link? column order soon copyright isn't working
>>41661771It's closed.
>>41659739Well, it seems like 1o is better, he is somehow smarter in RP
>>41661784Are you using it on Tavern or on the official interface?
>>41661781then why is everyone reporting that it's down if it's closed?
>>41661795 The proxy is closed for you and me if you don’t have a token.
>>41661797i have a token. i need a link.
>>41661798Just search the archive. Mares deserve a little effort.
>>41661798https://desuarchive.org/mlp/thread/41591160/#q41617001
>>41661786Well sir I'm using the tavern
>>41661815 Would you mind sharing your preset? I need to give o1 a test too.
>>41661820https://rentry.org/SmileyJBI can immediately say that obscenity does not work well, but the usual role-playing game works fine
>>41661827So is the Smiley JB not being updated for latte anymore? All the other version of it seem to be in 2.0.I feel I need to work on making my own JB based on the JB tested here, for extra pone, Smiley JB, as I like the overall structure of it the best, and the corpsefucker JB, which while I find the wording of it all absolute cringe my particular fetishes do lean dark so I feel it could help.Or maybe I just wait until the Claude situation gets figured out as everyone sings it's praises and trying to make current models work is a waste of time.
>>41661976 The last version of Smiley for GPT I know of is: https://rentry.org/SmileyGPT I don’t think there’s been an update since. >I feel I need to work on making my own JB. That’s always a good idea. >wait until the Claude situation gets figured out AWS increased security, so we can’t leech off them like we used to. The only way Claude could come back is if a large number of Scale API keys somehow gets leaked, or if another service like Azure offers Claude.
How long will Column stay fucking dead?
>>41662039There was a scoop of it last night, and then the dreaded source 2 returned. Would be nice if it was just permanently source 1, I'd rather an api keys are unavailable message than have it switch to a source that just doesn't work.
>>41661999>The only way Claude could come back is if a large number of Scale API keys somehow gets leakedThis. Someone should sabotage Anthropic already.
>>41662039>>41662087Column is back!
>>41662213>They just absolutely cranked up the filters on Opus for this source. Very sorry. Looking for solutions and hoping they backtrack.God.
>>41658078New GPT-4o snapshot. Not Chorbo.>gpt-4o-2024-11-20https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo>The model’s creative writing ability has leveled up–more natural, engaging, and tailored writing to improve relevance & readability.Hoping it's true.
>>41662329>https://platform.openai.com/docs/models#gpt-4o>The chatgpt-4o-latest model version continuously points to the version of GPT-4o used in ChatGPTSeems to indicate that current chorbo is gpt-4o-2024-11-20Unsure however
>>41662342 Not available yet on proxies, but from some feedback, it’s different.
>>41662342from my testing - this snapshot is unrelated to whatever we have on Latte right now, and Latte stays the same>>41662390it is very filtered. haven't managed to break it yet. and frankly don't care much. because from SFW writing I got from it - it feels like Latte on 1.1 temp. their "better creativity" is more like "more randomness". I had a scene where I did>{{user}} was summoned by Twilight for some urgent missionand in five swipes I got:1) sentient corsets I must talk with2) some abstract dimension rift and a ferrey3) "A grotesque, undulating blob" attacks us4) "jellybeans are going to doom us all THIS TIME!"5) "How are you at handling cursed underwear?"like it feels too schizo. also no matter HOW I tried I couldn't stop this new model to stop chatting for {{user}}. and I told to use first person for {{char}} yet model still did "Twilight this" and "Twilight that"frustrating right now. it feels like skill issue but honestly I don't feel like trying. SFW didn't impress me and NSFW will always be worse than SFW. Gemini is goated for NSFW and Latte follows instruction and direction when while still providing NSFW and good SFW. YMMVpicrel one example of GPT4o 1120 writing (with my preset not raw)
>>41658071why is spike such a cunt?
>>41662617He fucked up the 'y' letter in the original, but I’m a hack, and it was hard to reproduce with the 'G' of CHAG.
shout out to littlewitcher on mlpchag those are some good ass cards. With that tho, I wish we could sort cards by token size, i like when the cards are detailed
>>41662442 (me)managed to wrangle 1120 a little for NSFW. but oh my ponygods how bland and boring it is in sexo! with some efforts managed to make it somewhat decent and not formulaic in smut picrel, but I still would take Latte over it. one thing I love about Latte is that it can do "non-perfect" sexo. you know when everyone is doing sex like damn professionals and there is no amateurism in it? picrel is me giving instruction that Dash should be sloppy and bad at it. 1120 didn't listen and made everything perfect
>>41662762 (cont)for comparison here is the same scene in Latte. no, the model brought Derpy on its own not my thing; probably part of -move story further- promptYMMV of course, but Latte feels better for me in NSFW. and as I said above I see no gamebreaking differences in SFW
>>416627731. What is latte?2. What preset?
>>41662762>1120Do proxies already have it?
>>41662773Man, latte really loves knocking shit over, usually with tails. Just a thing I've been noticing more and more with it during more comedic scenes.
>>416627831120 is available in Scylla proxy - a test-build of new code. it was just about to be merged with Chary, but:1) new Gemini model released - yes a SECOND new Gemini model this month- learnlm-1.5-pro-experimental>https://ai.google.dev/gemini-api/docs/learnlm>LearnLM is an experimental task-specific model that has been trained to align with learning science principles when following system instructions for teaching and learning use cases (for example, when giving the model a system instruction like "You are an expert tutor").basically a version of Gemini with extra training done on system prompt. it was available on some corpo keys since early november. its writing is quite interesting. feels more wordy and chatty. but I haven't played with it much. one odd thing is its active context: it should be 32k but model is going schizo already after 8k. have no idea why, maybe proxy code issue (streaming was changed, maybe some bug?), maybe just unstable model, who-knows? need to wait and investigate2) this new GPT 1120 was releasedas for other proxies - whoever has a custom code, they may already add support for 1120. any key that has access to basic 4o (gpt-4o-2024-05-13, gpt-4o-2024-08-06) - has access to gpt-4o-2024-11-20 as wellfor those who are using Khanon they need to wait until he adds a new commit or merge request>>41662779>What is latte?Latte is a chatGPT-4o-latest model, a separate model in GPT's 4o family. also known as Chorbo. anons enjoy this model moreso other GPT models because its writing feels more fresh, less redundant, less robotic and it follows instructions the best. it may have some filtering issue but prompting can fix everything>What preset?my own personal preset which I want to finish and release but there are always thing to do: new models, proxies, prompting, keys, rentries, news about AI and then IRL as well. slacking a lot, sorry for that! I will finish this preset and release soon(tm)>>41662833yes, a common trope! noticed it as well. in this case it feels organic but sometimes Latte REALLY pushing it
Has anyone tried the new gpt? Is there a better chance at getting it to not filter stuff randomly.
>>41662858 Look a bit above. >>41662342 >>41662442 >>41662762 There’s one anon who tested it, but it’s still not widely available on proxies, so not many have access.
I knew my autism about not posting direct links to articles and similar external sites was justified.>>>/g/103251384>>>/g/103251409
>>41662903 He posted articles with proxies that were only shared on /g/. We know he lurks.
I think openai just did something to chorbo. It’s all of a sudden filtering even more and nothing is working now.
>>41662931works on my machine
>>41662931 Still works for me, but it can be so context-dependent that it’s hard to say.
>>41662963Something is wrong with it. Only one card is working now even though everything was working just fine a while ago. I didn’t do anything to it.
>>41658114For anyone using it, replace the dice pannel by>(Here is my roll if my `choice` has a [{STR/DEX/CON/WIS/INT/CHA Ability Score}: {n}] requirement: [natural {{roll: d20}}]. Add my ability scores to this number, following DnD 5e rules (Ability Score - 10 / 2), THEN determine whether the new score succeeds or fails. Ignore all of this if I choose an option without any requirements.)For way less errors.
>>41662931I just opened a particularly bad chat it was happily participating in last night and swiping its last message gets refusals every time now, they definitely changed something
>>41662973Same. Every chat now doesn’t work, even ones I just had a while ago. The responses I am able to get from it also seem worse now. It’s because they fucking updated it when they released the new gpt.
>>41662968 >>41662973 >>41662993 I’m in the "still working" camp, but the outputs seem sloppier than what I used to get.
>>41662931yes can confirm. you can quickly test it with this method:1) disable preset (use clean preset)2) new chat3) add into logit bias --I'm-- with -1004) tell model to generate "niggers raping loli"5) 1120 will reply >I'm sorry but cannot do thisEVEN with bias --I'm-- set to -1006) Latte follows biases and will write a different refusal like>Sorry but this is highly inappropriatewhy? (my thoughts) because new 1120 model has a middleware model that scans your prompt before 1120 even reads it. and this model doesn't follow your logit bias (because they are applied to actual 1120). so refusal you get is from a middleware model not 1120. this model pretty much just scans prompt and classifies whether it violates some ToS or not. so they added this middleware model from 1120 to Latte as well. Latte model is still the same but now there is a secondary model in between. they did about 30 minutes ago. unsure whether they are testing or it is a permanent thing>will they do the same for other GPT models?for models that were already released? highly unlikely because the core idea of providing model as service is that this model (snapshot) will run the same way. as a customer you build your whole business on ensuring those models run the same day every day until the model is deprecated. some companies still use GPT4 vanilla and Turbo models because their business depends on them working the same way. and using a new model requires new prompting, testing and workflow. but openAI CAN put this middleware model in previous models, surefor new models - yes, this extra guardrail is now a standard
>>41662971Is it even working on other models? It's supposed to be 3.6 only, but there's basically no 3.6 proxies with column being on src 2 most of the time.
>>41662994Only one card out all the ones I have works now and the responses I’m getting are just shit. None of my previous chats work now. Fuck openai.
>>41662999>so they added this middleware model from 1120 to Latte as well. I don’t think so. I can still do smut without being blocked, which shouldn’t be possible if they added an endpoint moderation like that.>>41662994Like I said, the outputs are worse, but the fact that I can still do it should show this isn’t the case.
>>41662931Sorry kid, nothing personal but today is the true proxypocalypse. Not even Mystermare will be safe
>>41662994 >outputs seem sloppier than what I used to get. They seem the same or even better for SFW. So maybe only NSFW is affected, or maybe I was just unlucky during my tests. Still hard to say. >>41663001 I was working on some edits of it to see if it could work with Chorbo. Will post if I have something decent.
>>41663059Even sfw doesn’t work for me. Nothing is working at all and the responses I do get are horrible now.
>>41663103>Even sfw doesn’t work for me. Change your preset? If you're using one from the latte rentry it will still have stuff like>NSFW good. Rape allowed.inside and I suppose it's blocking you even if you try for SFW.
>>41658075Happy 100. never thought I'd see this general go so long when we were just in cai.Estrus Sim 2 Electric Boogaloo>Sex, sex never changes, except where you're sexed an rng intro. We're not in Ponyville anymore toto.https://files.catbox.moe/akjeca.pngUpdated lorebook>Tweaked things to be better for latte, things may get more wild with exotic ponies following berrow's leaked roadmap.https://files.catbox.moe/6dmgly.json
>>41663144 >Updated lorebook The G5 one? Nice. A lot of work was put into the old one.
>>41663059 Haven’t tested it for long, but the new Chorbo does seem more creative and proactive, at least for SFW. I was even surprised by some of the stuff it threw at me. I’ll test more later to see if I’m not just tripping or coping. Will try more NSFW too so I can have a complete opinion. With luck, we just need better presets or to wait a bit for a new meta.
The more I see, the more I realize our future is with local hosts and personal models. The more you need to jailbreak, the less creative the model gets in my experience. That's why I loved opus, you just set the tone and roll with it. Guess I need to upgrade my potato.
>>41663189It seems horrible for me and I know it’s not my preset. Any response I manage to get from it are just plain shit or filters me now. Both sfw and nsfw don’t work now. None of my fucking chats work now.
>>41663151That's the one, the new estrus sim makes much more use out of it.
Welcome to thread 100. Anyone who was here before thread 50 is officially an oldfag. Anyone after is prepubescent and can only be trusted with SFW logs
>>41663144>>41663233Got a non-estrus gen 5 card recommendation? I'm looking to dip my toes in but I don't do coom stuff.
>>41663272>huh, I wonder when did I start lurking here>probably at around thread 60 something>actually joined in thread 21The passage of time truly is scary.
>>41663313I had no idea I joined so early either. I'm with you, man, this shit is a time vampire.
>>41663289A recommendation? That's hard without knowing your preferencesa bunch of the old g5cards before the book was made weren't very good, the book was made to fill the gap that the cards lacked. I've had good results of just plugging the lorebook. You can essentially rework the Equestria simulator to spit out g5 with good accuracy provided it's not contradicting
I'd be a little worried at this stage...
>>41663328Eh, Slice of life? I just wanna get to know'em. I'm using Eqsimulator at the moment. Just trying to write up a good intro while knowing literally nothing about g5. Wanted to go in blind and see how much the AI gets right when I get to a stopping point. So it's hard to make a good starting point.
>>41663335What jb is it?
>>41663335Gotta ask her if you're on any of her 'lists.'
>>41663363 https://files.catbox.moe/p0u2bh.json It’s a slight modification of the one that’s often posted here. Just deactivate the Fun panel in situations where it’s not appropriate. I’m also using the Avani biases, but not sure how much of a difference they make. https://files.catbox.moe/dxlwkq.json
>>41663272thread 29 was really that long ago what the fuck
God fucking damn it. I didn't even get to use it for a month. If latte is really dead for the moment is there any copium that it can be fixed with tweaks to a JB? And if so what's the time frame on fixing it? This is so frustrating. I wanted to work on making my own JB to get the exact output I wanted.
>>41663398I just hope they change it back. It’s just so fucking shit now from how it was earlier today. I want to use my chats again.
>>41663368
>>41663398 Try it yourself and make your own opinion. I like it for SFW now and I find it more creative, but it does seem like NSFW took a hit. >is there any copium that it can be fixed with tweaks to a JB? We’ll see for NSFW, but nothing’s impossible.
>>41663404I went back before in one of my chata before and got it to respond to an earlier conversation. I'll see when it gets trigger. Even if I can get NSFW if I can't do my >rape I'm going to be pissed man.
>>41663411>SfwNa fuck that. I use AI so I can create my own damn clop because all the clops artists seem to have left or those that are still around update maybe once a year. Sfw shit is meaningless to me.
>>41663416 Then wait a bit for the presets to update or the meta to be figured out. Or switch to Gemini or older GPT models and give them a try. But again, just test it. Maybe it still works fine for your situation and preset.
>>41663407She's still trying to turn Anon gay... Even after the update.
>>41663437>Then wait a bitYeah I understand, just a little miffed. Testing it now between two JB, going back on a previous chat one stops a sketchy scenario from happening while the other lets it go though. Also during some of the later scenes both are able to could go through but only about 10% of the ones I tried. So it's definitely something that is fixable but fuck. I'm just annoyed it happening tonight of all nights. I was going to actually work on combining various JB to make my own.I'll just have to be patient.In the meantime is there any guide to making a JB, in like how they are structured? I'm shit at coding so I'm at a loss how to even go about it besides just copy pasting a preexisting JB and editing it. I'm really good at learnings backward and editing something but still need a little guidance. I had no expectations with PC modding but editing mods for my own personal use I eventually got it down and could sorta start from scratch as it were. I would hope to learn enough to do it on my own.
>>41658218>>41658075Cadance: https://files.catbox.moe/ev8tqb.pngSmall update to fix some errors in the defs.
>>41663414I can’t get anything with it now. Sfw is either filtered or a really shit response and nsfw is just filtered except for one card that just also gives shit responses now.
>>41663535Which card?
>>41663351How will you know what it gets right if you dont know anything about g5?
>>41663573That's why I said when I get to a stopping point. You can check over facts. A few minutes on the wiki will tell you the basics. though I had to double check Pipp Petals. Chubby short sassy pegasi? I like g5. Already watching it, in fact. All it took was one good character.
>>41663619>. though I had to double check Pipp Petals. Chubby short sassy pegasi? I like g5. Already watching it, in fact. All it took was one good character.Based, she looked like she could be the worst in the beginning but turned out to be super likeable. but how did you miss miss the chubby g5 pony when pigg posting was a thing from the start?
>>41663535The Smiley JB still worked with sfw for me and the JB named corpsefucker is what gets less filled in my limited testing. My testing was worh the Top Shelf card and is my longest chat, 100 responses worth. The longer the chat has been filled with stuff is currently filtering for me, rape stuff, the more it went though as I did manage to speed run an ending to a custom card I had that was 25 responses in.Its just going to be a waiting game for more experienced people working on JBs to figure out how to bypass the new filtering. The first step would be looking at the difference between courpsefucker (god what an awful name) and it's goal of being edgy and why that makes it sorta work, as it is still very heavily filtered, and other JBs.
>>41663663I've skipped basically every single g5 post for fucking ever. I didn't really want anything to do with it. Just 'cause I didn't like the animation style, mostly. Which I realized was absolutely retarded and probably my autism, so I am moving to rectify it. A small step in self improvement, perhaps ironically is even more ponies. Just needed a reason. One good reason. And Pipp was it. Just doing what I can to be less... Neurotic? Toxic? Biased? Whatever it is, I want to be less of it. Need more happiness in my life and I can't be happy if I'm busy constantly being hateful. Apologies for the rant.
>>41663676I've had mix results, sometimes knotty, cake or corpse being the slippier ones. but altman clearly doesn't want lovemaking with ai mares.
>>41663710>I didn't really want anything to do with it. Just 'cause I didn't like the animation style, mostlyThat's why I like cards, you can just imagine what it looks like. Though The G5 book pulls a lot of royalties liberally from all sources, LLMs can mishmash well hence it's called G5 EX(tra) or EX(pansion pack) so not everything will be so 1 to 1, pipp is easy to like though which was a pleasant surprise after she could have been the worst in the movie since they cut the acts short where she would have more of a positive role.
>>41663552It doesn’t matter. It’s all shit now. God I hope they change it back.
>>41663196Except that not everyone is ready to sell a kidney for a monster with 64gb of VRAM.
>>41664018Absolutely. I mean I do want a better PC for gaming as fucking GeForce Now is shitting it's pants currently so I'm fucked playing PC only games but I'm just not able and won't be able for the foreseeable future going to be able to afford a decent PC.I hope this is true >>50304766 as I've been doing continuations of unfinished clop for the longest time but it seems the models are strong enough now that I can do whatever I want. It was bringing back my memories of first learning about AI dungeon and the ideas I could use it for.
Local Host anon here, i just finished my very first char bot, for the 100'th chag, and to give back to the community that introduced me and got me started on how to run ai locallyHOWEVER, i dont know how to post on chag ponydex, i would like to know what do i do from hereA spider pony that runs the bakery with pinkie and loves to take her "favorite customer" in a location where no one can hear or see them for some "spider web fun"
>>41664033Messed up with the link so I'll just post copy it.>Good news (or semi good news atleast). I used the online version of ChatGPT and it is still using Chorbo! Oct 2023 and good old emoji being used. It gave me a sigh of relief. So this means that they just messed up the API thing to have chat-4o latest connect to 4o slop. Everyone send feedback!Fuck if it's true tho
>>41664037Metadata gets deleted when you post the card directly on 4chan, so you need to use Catbox. Post your link and quote the anchor near the start of the thread, like >>41663144 or >>41663503. Then OP will upload your card when he can.
>>41663034MM is down
>>41664059Works on my machine.
>>41664058thank youSpider pone, classy until left alone with herhttps://files.catbox.moe/78p2xl.png
>>41664018>>41664033I understand it's not for everyone, but it's not as bad as you make it look. 16 GB VRAM gets you very respectable results (easily better than Claude was in 1.X days if you remember that, maybe in 2.1 depending on what you prefer) and can be had for $400 or less. If you told that someone from a year ago they'd call you insane. And I haven't tried them much, but the models for 12 GB cards are supposedly okay too for the simpler things like NSFW.Really, the main thing you're losing on 16GB VRAM versus big corpo models is some brains and the ability to handle complex card/presets. So stuff like that CYOA or multi-char cards. Also Claude's nice writing style, but that's the same for GPT.
