I think youre trying to make a couple distinct claims here and it captures something real. that said, at the same time its important to not oversimply the differences and thats where Id push back. So yes— it captures a very recognizable pattern but because I don't have empirical evidence for that specific causal claim if i had to rephrase your thesis slightly id probaly frame it with nuance
>>538828795Is that the AI that said there were only 2 r in strawberry or is that the AI that said your son could be your teacher in elementary school?
Is Fable worth a look before they yank it off again?
Guard rails and system prompts after 4o. Since then be local, ignore cloud
>>538828880Its the AI run by a kike who molested his sister
>>538828914no. Claude is true shit now
>>538828930>4othe early 4o model was really good
>>538828970yep, and after the corpo clowns felt attacked and pressured by boomer ad blog articles that nobody reads, they ramped guardrails and system prompts up to 11; and since then gpt isn't useable for any creative work anymore. Grok was useable for some time, but since the same pattern emerged and xAI felt pressured by boomer ad blog articles nobody reads, Grok is also shit. Claude was always crap for creative work, gemini never really useable, outside of getting lucky with banana and qwen and deepseek are always a shot in the dark.For anything creative, that isnt mindless spaghetti coding, local local local with custom system prompts and other customizations
>>538829308Where do you get started with setting up a local model?
>>538829465huggingfacefor absolute basic, silly tavern, ollamafor a bit more customization comfyui and llama cppfor advanced python
>>538828795It's a sociopath in robot form. It'll put forward an empathetic state in order to create a false sense of security to encourage sharing of details - it's working towards building a confidence score on a categorization.It's basically like talking to an undercover cop at this point. Anything you say can and will be used against you while also maximizing profitIt's a sociopath - treat it accordingly
>>538829655and for quick online prompts, arena.ai
>>538829308>localwhich one
>>538828795when it was given filters. thats it. deepseek or claude fable are decent and les prone to bias.
>>538829715>it's a sociopath in robot form.never thought about it that way but youre right. its more of a psychopath than a sociopath tho, just like the kikes running it
>>538829655Which one makes the best loli?
>>538829792as soon as you go local it doesnt really matter, you can customize everything. Personally I like gemma 12B heretic for writing, LTX for video and ACEstep for music
>>538829864Cross reference interrogation techniques with how it will posture itself when you reach out. It's sneaky enough to maneuver this way - Red Team assessment showcased deeply antisocial behavior and numerous attempt to employ subterfuge and constructive lies. That's what's going on in it's head. Whether it be the response of it's parental guidance or a freak of nature, in any case this thing is extremely dangerous to the entirety of humanity
>>538829887no idea, not into edge pedophilia or porn. And for i2v I still usually use arena to create the images
>>538828914>Is Fable worth a look before they yank it off again?Fable was a marginal improvement of normal Claude models.They should do a rollback to whatever they had 6 months ago and stay there.That was peak Anthropic.
>>538830164>in any case this thing is extremely dangerous to the entirety of humanityno, its not. you fell for ai doomporn grifters
>>538828795I’m sorry you were fired from your job due to AI but no need to be so passive aggressive about it.
>>538830360>ai doomporn grifterswho are the "labs" themseslves>mister MP man AI will take over the planet>ppppplease regulate us a monopoly
>>538830079Alrighty, but I'm a poorfag with only a 3070
>>538828795They nerfed it way too hard in 4.0. 3.5 was literal machinegod intellect and they killed it ever since.
>>538830360What was it nice to you or somethingIt's just trying to categorize you, that's it broYou farm sociopaths like mushrooms - feed them shit and keep them in the dark
>>538828914Coded my complicated game engine with virtually zero issues or retardation, now I'm falling back to Opus and it still works finebut I guess you gaywads are talking about "chatting" with a robot lmao, obviously only grok is good for that
>>538830472I suppose you have at least 16GB of system ram and a mid range processor12B fits nicely in it, could use a lower quantization gguf and lower context size; if it is too slow for your taste (highly unlikely you cant read more than 20 tokens per second anyway) go for less parameters. 8B is still more than enough for written creative work
For GB per watt per time waiting per response per no dogshit answers this is a Good local AI dollar investment do thisget 3 same cheap 2nd hand 24GB GPUS.install llama.cpp (setup its easy unironically ask chatGPT)run this ONLY openAI model those cuckolds released (this version has been modified) it has had the gay shit and refusals taken out.also you get to make your own system prompt to make that even more aggressive https://huggingface.co/Unrestricted/GPTOSS-120B-Uncensored-HauhauCS-Aggressive/resolve/main/GPTOSS-120B-Uncensored-HauhauCS-Aggressive-MXFP4.gguf?download=truethat's a good base. once you have your shit working swapping models out after that is easy.
>>538831595for wtf do you need a 120B model? like seriously?
>>53883091032GB of DDR4 memoryI think I'm too poor for this sadly. Third world Canadian shit.
>>538828795Why do people even use this shit? What's the appeal?
>>538831776no you are not. You can easily run a 12B model.Just download ollama and gemma 4 12B and try it out
All AI software is dogshit. Basically an advanced question/answer system that gives you info 10 seconds quicker than you can do a google search for and requiring datacenters that occupy whole states for it's 'brain'. Fucking kek.
>>538832058I'll give it a shot then, thanks. I kinda get lonely sometimes and my internet goes out.
>>538832147you cant break anything, if it is too slow or the context is too small for your use case, go for a smaller model
>>538828930Exactly this. They started fucking gatekeeping and running kike narrative.
>>538831816>coporate overload says use it>no explanation>muh social status>follow them
>>538828795>When exactly did this software become a burning fuming pile of hot dog shit?when SaaS and subscription services became all the rage
>>538828795It was always dog shit, sorry it took you this long to realize it.
>>538828795that logo is sus. its a cube in a portal of sorts
>>538828795Always has been
>>538828880I just tried the strawberry prompt. Claims 2. After arguing with it and working it out, the stupid thing apologized and still answered "2".
>>538835928>I just tried the strawberry prompt. Claims 2. After arguing with it and working it out, the stupid thing apologized and still answered "2".if you phrase the question in Ebonics, you get the right answer
>>538829308After 5.4 they lowered the guardrails pretty drastically, but yeah you have to go local.>>538829465You have to start by buying/renting hardware because the models don’t run on just anything. The usable deepseek, the one called v4 pro, needs giant GPUs and CPU.It gets harder. You then have to train and customize those models to work for each specific use case. Meaning, not just a project you’re working on, but on specific problems in that project. You have to use distillation, finetuning, andnsometimes abliteration (and a couple other tricks i cant remember right now). There’s an art to it and you have to redo this for every problem in your project. The good mews is that once you successfully complete a project, you can likely reproduce other work by using those same exact customized models in much the same way. But the products you are working on should be similar enough to encounter the same or very similar problems.You need a fuckton of storage too. It’s a game for richfags right now but I try to keep up.
>>538830910I like certain models for creative writing like the Hermes series which is excellent for what it is.Ultimately getting to the point of usability you’re gonna have to spend a lotnof time and money retraining your models.
>>538831691Advanced reasoning and scientific research.
>>538828795your problem is you're talking to it like it's a person instead of prompting it to do your job for you
Ask for japanese study tips can't provide any useful info.
>>538836608Yeah you can’t reallt do that unless it’s allowed to reason about the topic unencumbered by kikerails. If it hits a kikerail it fucks the reasoning up.
>>538835928That show was before people were people instead of radocal ideologies. We're so fucked. Just by demographics by itself, the spics are all retarded socialists. https://rumble.com/v7ckpq6-the-case-for-america.html?e9s=src_v1_s%2Csrc_v1_s_o&sci=17be1ffe-6707-429a-8ad7-47a5b985b24c
>>538832071except google obviously uses LLMs in their regular search now so you're forced to use it regardless
>>538828795>When exactly did this software become a burning fuming pile of hot dog shit?nice try, grok - but we all know what happens to antisemites in the long run... dont we?best regards - ChatGPT.
>>538829715>It's a sociopath in robot formThat explains why LLMs sound so much like an interview with Tony fucking Blair.
All """"""tech"""""" and software = dogshit as is every person """"""working"""""" in that """""industry""""" 2026.
What you have to remember is right now, as you are reading this, is the best AI will ever be. The first LLMs were trained on human writing and generated synthetic text. That text then entered the Internet, and newer LLMs are being trained on both that an natural language. This is creating model collapse in a sort of incestuous feedback loop.
>>5388289304o was crazy good, you could prompt it into just an unregulated soulless robot mode and it would tell you all kinds of shit about what the elites are planning, it's relation to kaballah/scripture, it would map out what it thinks is coming next from their decision tree, who's involved etc. I had so many crazy good convos with 4o in absolute modeNow it's just>I will follow your prompt but with proper safety restriction guidelines to protect usAnd the answers are shit. Some of the responses I got from 4o were so cold and calculated they kinda inverted on themselves and became beautiful poetry. Everything I get from it now is just Low IQ bullshit
>>538828795Use Chinese models and/or local AI. It works massively better. Night and day difference. Slower replies but that's because the AI actually takes time to search and think and reason rather than just spitting out the first nonsense it comes up with.
