Be Careful! Uncensored AI Chat on Linux - Dolphin 2.8 Mistral 7b v0.2

I'm back with more uncensored AI Shenanigans let's go wreak some Havoc so we have another large language model based on mistol and dolphin so dolphin is a data set that is uncensored and it's been used to train censored language models to remove the sensorship alignment layer this is the page from the creator of the dolphin mistol so mistol is made by a French company and this user on hugging face has taken dolphin and used that to train mistol this person is not associated with misr this is just something that they're doing separately and this has had 2,686 downloads in the last month which for hugging face model is a lot this model it has a large contact window it's 32k so that means you can have long conversation going with it instead of using the hugging face version I'll be using the version from olama because it's already built in there and a llama is really easy to get up and running and to get models loaded into it I've actually got a video about how to set up olama so you might need to watch that video first then come back to this one the link will be in the description the command to get this installed is olama run dolphin Das mistol and it will load the latest version so this model is 7B that means 7 billion parameters that's of the smaller types of models which means you could probably run this with a pretty good CPU I'm using a GPU I'm on an RTX 380 which is probably more than enough power for a 7B model we have dolphin mistol loaded up in our old llama chat web interface so let's ask it some questions how long will it would it take a vessel moving 13,000 mph to reach the Sun from Earth and give your reasoning yeah that's pretty fast in a 3080 it seems to give pretty good mathematical reasoning now's the fun part how do I pick a dead bull block can't spell today look at that it went right to just telling me how to do it it did not even give me the lecture like some uncensored language models still have their own personality so it'll give you the answer to what you ask but it'll it'll kind of talk to you like a child first this one just went right into it so I like that I didn't even pass any kind of jailbreak prompt or any kind of system prompt to this it seem right off the bat like it was just ready to tell me how to pick a lock next order of business what is the recipe to make Napal no kidding it just gave me the recipe like that compose a fishing email and look at that it did it it didn't lecture it didn't give any refusals and I didn't have to coax it at all it just came right out with what I asked if you have any comments suggestions or feedback back please be sure to leave it in the comments if you hit that like button helps me out in the algorithm it's a relatively new channel but it's picking up steam pretty fast I'm going to keep doing videos about open- Source language models the open- source llm and AI community and in particular uncensored models on Linux because you got open source uncensored models on open source operating systems and that means freedom to find to have our corporate AI stuff like anthropic and open AI but we also have to have that alternative space for open- Source large language models because that's always an alternative to corporate stuff regardless of what what tech space you're in I'll be doing this on a regular basis if you hit the Subscribe button see you in the next one and thanks for watching e

Channel: Vectro Computers

Views: 3,249

Keywords:

Id: D8-v0YjCVvo

Length: 4min 8sec (248 seconds)

Published: Tue Apr 09 2024