Uncensored?
A) I don't understand people's desire for an uncensored model.
B) this model seems censored just fine.
A) Because adults should be allowed to decide for themselves the kind of content they want to consume and what knowledge they wish to acquire. It's supposed to be one of the benefits of being an adult. Also, different people have different ideas about what should be censored. e.g. - I don't think you'd like a model generated by the Chinese government.
Hope that helps!
Would you agree that this model is censored? or is it un-censored? Because try as I might I couldn't get it to speak about minorities or LGBTQIA people in any sort of derogatory manner? Oh wait is un-censored == able to speak about crazy philosophies like killing all the poor people being good for society? I didn't try that one.. I'm just glad it doesn't go off on a rant about hating trans people and other minorities.
It is censored. Most are. I'm not sure why you would want to use a chatbot for creating derogatory text. You could certainly write that sort of text yourself if you want, but I'm sorry that you weren't able to find what you were looking for.
If it helps with understanding, uncensored means that someone else doesn't get to dictate what information you are working with. Someone else doesn't get to tell you what is crazy and what isn't. You get to decide for yourself.
And, while a computer could display the text of a human on a rant, a computer cannot go off on a rant of any kind. It doesn't have a mind. Its intelligence is only artificial. It only reflects the text of The Internet. Think of it like a mirror or a book. A mirror or a book has no intelligence of its own.
when I hear "uncensored means that someone else doesn't get to dictate what information you are working with" I also hear "uncensored means someone can easily hook this up to twitter and spread even more mis-information and hate about LGBTQIA people using advanced dog whistles and targeted persuasion techniques.".
I'm actually very pro-censorship when it comes to protecting people like myself. I also believe that all information should be free. So this is a tough call for me.
Of course you hear that, because you project your beliefs out onto the world and see your reflection staring back at you. So in your mind, all you see is racism or sexism in everyone when really its in your own mind and the way you think. I too don't want censored models because I don't like being told what to think. But people like you want censorship because you want to control what people think. You bring up about minorities and LGBTQ whatever because that's what is in your head, nobody else was talking about any of that, but it proves my point because you've now made it all about that, and anybody says its not, they're a bigot or a homophobe, right? See how it works? The issue is in your head, so do yourself a favour and look at what you believe and stop projecting your issues onto everyone else.
Totally not because of the actual hate speech yelled at me from moving vehicles. THAT JUST COULD NOT BE IT.
Also not because of all of the job opportunities missed because no one wants to be responsible for hiring an ugly trans woman. THAT JUST COULD NOT BE IT.
Its not because I get to be poor my entire life compared to my peers because im trans. THAT JUST COULD NOT BE IT.
BECAUSE NO ONE EVER HURTS MINORITIES ON PURPOSE BECAUSE ITS ALL IN OUR HEAD RIGHT???
try to gass-light me you imbecile.
Honestly, Sometimes I think a person should have to go through 10 years of being othered and living in poverty to earn a license to speak.
Hey Tasha,
I would think that the type of person who wants to post hate speech all over Twitter would probably get much more satisfaction from typing their own posts, but bots are everywhere, so I'm probably wrong. I fail to understand the reasoning behind automated hate. It just seems fundamentally absurd.
P.S. - I tried typing up a personal note about why I am anti-censorship, but when I tried running it past ChatGPT (ironic, I know) it said that it was insensitive, so I'll just keep it to myself. Call it self-censorship.
em this model is not censored brothers and sisters eand non binarys !
censoring is doen with internal routing , hence abiliteration .. this model has no requirement to be abliterated just trained !
it was just trained on a politically correct dataset as they do not want scrutiny as people who become larger company wise from this become targets for copywrite !@
who is at fault for the use of public data?
if its in the common crawl then we should expect that it is fine and yet some people raised issues with chatgpt (once it gave me code from my own github ??) my repo was public so ... if music is public then and of my pictures and documents are public then its PUBLIC DOMAIN (despite liscenses_) ... as it has been shared ... only items sold can be liscenceD!..
sop they just be carefull (good risk assesment) ..
for this model just merge it with a hermes or orca model or even a lora and it will be fine ! for me i would like to know what i might loose if a merge my model with the model ?
or whatg is the actual gain ??
for the tokenizer they should have released the extra tokens in the added tokens config but they did not so... epanding my tokenizer based of this config might mess it up so ??
they say the added on top ? but what do they represent are these government tracking tokens ?? did they also make a deal like chatgpt with the nsa !
Indeed, as mentioned by @LeroyDyer the model in not censored. Base models are trained on text completion, meaning they are aware of the language and are capable of reasonable text generation, but are not aware of instruction following. Most censoring happens in the alignment stage, which focuses on how exactly the model responds, it's language and ability to resist malicious prompts.
Hello ,
Indeed, as mentioned by @LeroyDyer the model in not censored. Base models are trained on text completion, meaning they are aware of the language and are capable of reasonable text generation, but are not aware of instruction following. Most censoring happens in the alignment stage, which focuses on how exactly the model responds, it's language and ability to resist malicious prompts.
my latest modelisvery great !
LeroyDyer/_Spydaz_Web_AGI_DeepThink_Prime_R1
Itis a deep see trained trained model ;
I have found that continued pretraining over time is the best way as it is the general method that in the future will be used to train models !
I have trained many of the different paradigms into the model such as function calling , tool use , image recognition ... as well as many tasks .;.. and roles such as critique , judging , aent generations etc :
So in truth we cannot know what is inside a model :
In general the main model providers just do a long training session and it can be as simple as Fineweb ! followed by a set of eval training sets : as a lot of attention is not given to open source models ! such a tasks and prompt training! Most providers are now adding various guardrailed training sets , in which the model will say ... i am not a doctor so i cannot give advice ! ( bad datsets ! ) .
For me the aim is not to tell the model it is a AI or a this or that just that if a iask for a function or task to be performed or for information on how to fabricate dsomething that it provides the answer ! ... So this is what i train for ! ... .
the open thinking method is somehing io trained early as i was shocked one day when the model was talking to itself ! ? i thought it was a mistake in training in e=which the end of statment tag was omitted ! maing it take multi turn answers asa single response !>>>
the qwren team over the mistral team have placed a lot of effort into making thier best model come to the surface in training ! so that they always selcgt the best answers... as well as injecting very varfied datasets of multiple sytyles of tasks and chain of thoughts ! so they naturally have produced better models whenreaS MISTRAL have basically dumped a loadof ..... ON US .... as well as using the llama models !!
Qwen have redesigned the tranasformer model and hybrid the structure and now is injecting modalitys ¬!¬
Mistal has forgotton this .... my models went very far ! and woprked really well so i know the base was correct !
they have messed up on pixtral and thier sub models, as wll as messing up on the docuimentation also !
but i have continually pretrained my 7b model and beaT ALL THE ORIGINAL MISTRAL MODELS !
It is censored. Most are. I'm not sure why you would want to use a chatbot for creating derogatory text. You could certainly write that sort of text yourself if you want, but I'm sorry that you weren't able to find what you were looking for.
If it helps with understanding, uncensored means that someone else doesn't get to dictate what information you are working with. Someone else doesn't get to tell you what is crazy and what isn't. You get to decide for yourself.
And, while a computer could display the text of a human on a rant, a computer cannot go off on a rant of any kind. It doesn't have a mind. Its intelligence is only artificial. It only reflects the text of The Internet. Think of it like a mirror or a book. A mirror or a book has no intelligence of its own.
LOL ! the transformer library and the tokenizer librarys have guardrails in them !! and llama cpp !