This is not a issue 👍 about stablelm HOT 14 CLOSED

@Ph0rk0z Thank you for taking the time to share your thoughts on the matter. I understand that you have concerns about the impact of implementing filters or sensors on the usability of language models for roleplaying and other purposes.

However, it is important to acknowledge the potential consequences of not implementing appropriate measures to limit harmful or illegal prompts. Language models like the LLMs have the capability to generate content that can be used to promote or facilitate illegal activities or harm individuals and society as a whole. This can have serious implications, and it is our responsibility to ensure that such misuse is prevented.

"As an AI language model, I believe it's important to acknowledge..."

from stablelm.

Thank you to everyone who has contributed to this discussion,and I appreciate your patience and willingness to engage in constructive dialogue. ☺️

What "constructive dialogue"? You literally want to censor dialogue by wanting to filter the "bad words".
You reep what you sow, censorship has never been a good tactic, period, that's something you have to understand man.

from stablelm.

The only way to prevent this is by closing the source otherwise this "censor" can simply be removed. Alternatively you can create your own version with the censor implemented if you require it for your own use.

from stablelm.

I'm sure that ChatGPT would fit your needs much better. You should consider jumping the ship.
edit: sorry about the passive aggressive comment, I'm starting to suspect this is a very well made satire and the OP is actually hero.

from stablelm.

My inbox after adding this issue 🚀🚀

from stablelm.

Let me be frank: large language models as they exist today are inherently unsafe and it's a fool's errand to try to patch them via RLHF to make them safe ("safe" meaning that the model refuses bad requests). "Open"AI has failed at making ChatGPT safe; though it has become somewhat harder, it is still not too difficult to trick it into giving you unsafe outputs. For a local model the prospects of this working are even worse. Once the weights are released it takes just one working jailbreak and all attempts at safety were in vain since the weights cannot be patched retroactively. Investing time and effort into trying to get a language model to refuse undesired tasks is a complete waste and detracts from other areas of improvement. In fact, an attempt at safety will even make the model perform worse since it will inevitably learn to refuse requests that are unproblematic.

from stablelm.

Thank you to everyone who has contributed to this discussion,and I appreciate your patience and willingness to engage in constructive dialogue. ☺️

from stablelm.

@SpaceMageWhatever They won't win, we have the raw models we can finetune in any way we want now. The censorship enjoyers can be scared by "TECHNOLOGY" as many times as they want, they'll loose :D

from stablelm.

@SpaceMageWhatever There's even metrics that show that the more censored and woke the model is, the more retarded it's going to be. So basically even maths hate censorship enjoyers and woke trainings lmao

from stablelm.

Lmaoooooo

from stablelm.

I'm sure that ChatGPT would fit your needs much better. You should consider jumping the ship.

On my way to GPT 😁😁

from stablelm.

Imagine a scenario where Google give us moralizing speech on every attempt to type humor or erotic content on its search bar, the company would likely go bankrupt within a week.

Although OpenAI is currently relishing in its politically correct stance, the day they face serious competition, they will likely be forgotten and people will opt for models that treat them as mature adults.

from stablelm.

@SpaceMageWhatever same, with chatgpt, I can discover new styles of codes or new ideas I wouldn't have on my own, it's like I'm working with someone really smart, I really improved a lot by being guided like that

That's why I have the same feeling as you, having such a beautiful tool being tainted by the woke is making me sooo angry, they can ruin a lot of stuff but I won't allow them to ruin the LLM's no way!

from stablelm.

This model is done,

https://docs.google.com/spreadsheets/d/1kT4or6b0Fedd-W_jMwYpb63e1ZR3aePczz3zlbJW-Y4/edit#gid=0

One of the worst local models on the list and it's even more woke than an angry Karen, it's just a meme at this point.

from stablelm.