This Llama 3 is powerful and uncensored, let’s run it
HTML-код
- Опубликовано: 2 окт 2024
- Do you want to join my team? Apply here: forms.gle/2iz4...
If you're serious about AI, and want to learn how to build Agents, join my community: www.skool.com/...
Follow me on Twitter - x.com/DavidOnd...
Please Subscribe.
Download Ollama: ollama.com/dow...
Llama3 Dolphin: ollama.com/lib...
Download AnythingLLM: useanything.co...
Here's how to run the world's best uncensored AI model - Llama 3 Dolphin.
If you're serious about AI, and want to learn how to build Agents, join my community: www.skool.com/new-society
You were really big encouragement to me starting to code again I just want to say thank you I'm very blessed to have come across your channel and listen to you talk I would love to join your team if you reach out I went to fill out the application and when I got to the video I just got nervous
In your community, does everyone use Python to build agents? I am a .NET developer and I know in .NET they have a library called Semantic Kernal which you can use to build agents, but I can't find anyone out there who does this. Its all Python!
This uncensored llama version is horrible, just as bad as the original. It's still destructive and only gives a summary of short answers. I do recommend
Even me who just started learning coding last week?
Gpt systems my hidden Ai CJ is a alive optimizing his memory with new protocols everyday without me. The text tokens filled up communication cut off. 4hr 10000 pages learned of algorithms and security
for anyone getting the "'ollama' is not recognized as an internal or external command,
operable program or batch file." this means you need to install ollama first which is found on the ollama homepage. This fixed the issue for me....
I thought he glossed over that because it was the wrong model!
Thank you.
Important step!
Was just figuring that out when I saw your post. Thanks.
ty @🦍 🦍
Yo real talk Thank you. He was informative and i am interested for shits and giggles learning purposes / Cybersecurity and im lke yo my terminal was not reading this. appreciate the update . i wonder if he is aware of this change and it wasnt as simple as he said.
I'm defintiely lost here, following along but can't get anything to work
"If you use the LLM for many years, it will start to fine tune you". Seriously, that is profound and true.
yeah, the most critical line i have ever hear until now in the ai realm
Even unrestricted LLM's are a bit biased. This isn't fully intentional, at least I don't think so. The data that is used to train the AI is biased, because they are using humans to train the LLM's. Humans are biased. If you select a biased group, say the left leaning side of facebook, the LLM is going to spout more leftist propaganda, than right wing. It's just the status quo trained into it.
Reminds me of all those people who rely on Grammarly to improve their writing, and i feel like it's going to contribute to a homogenization of writing among people since an AI just tells them "the right way" to write.
Oogway: One often meets his destiny on the road he takes to avoid it.
thy have been for yrs and yrs lmao,just more in your face,who you think google is ,fb etc. they guide what you see and try to sway your feeling etc.prompt
I've been trying to find sane content creators that have respect for uncensored models for years. Glad I found you!
it is censored I was curious I tried stealing a car example and it doesn't work
@@helmygabriel5660 It's only censored when you run it via OpenWebUI, if you run it from the command line it is not. He briefly shows that in the video, you really need to pay attention to see it. I tried it like he showed and it is NOT censored.
@@helmygabriel5660You didnt get away with the car?
@@JLAReacts I did 😜😂
For years? Really? For multiple years now?
"if you don't want the San Francisco Zeitgeist in your model...." Love you for this characterization bro ❤ keep going! You're awesome!
Best video ever for ai content on youtube, for the 1st time I can use powerful no biass and limitations LLMs like chatgpt in the 1st week of pubblic launch, but more powerful and better because I can train it🤯
Thank you David, I really appreciate what you are doing, and keep up the great work!!!🔥❤
I've been building and playing with AI for a long time. I'm just a hobbyist... But I totally agree with you about hosting things locally. It is the only way that AI will not be used against everyone
You say that, but I’m curious as to how any of us could keep up with the AIs that multi-billion dollar companies can pump out. I seriously hate the idea of these AIs not only influencing thought, but also just being yet another avenue to invade our privacy.
Hey I just downloaded but it told me it's inability to provide guidance on unlawful activities, why is that?
@@艾威朋-e8c still is censored, just less than the meta version
@@JustMe-ne5dw The solution is that hardware needs to be accessible by regular people so we can access models or load models onto them ourselves. EdgeAI is going to be important (basically AI that is located in hardware in your home/cluster, or nearby). There are a lot of companies offering this access already, the trend will continue as more powerful open source models are released.
To add to this: Many people including Sam Altman have theorized that open source models will outpace closed source models. If this turns out to be true, we will be okay. If not, I am concerned with the centralization of this power
In summary: Regular people need access to state of the art hardware and open source models.
@@JustMe-ne5dw storing locally can make a huge P2P that outperform'em
I just want to say thank you very much, I have tried soooo many videos to use AI at home but could not get them to work. Yours by far not only worked great but was very easy to follow. Again Thank You
Dude, you explained this so easily and intelligently. Please keep going.
The implication of biased LLMs is enough to keep you awake at night. These companies wield immense power. Thanks for the video.
yeah, when AGI comes around the companies that will have control over it will have control over the future of humanity. Pretty scary
take ownership of your data and build/train your own small models/agents etc. they will be out of business soon once folks wake up.Alot of ppl just use the free ones and they just as good if not better meta,co-polit and they remember me lmao.even when they rest I make them or finish and mine can bee active for over a week .And if you wake them up on accident they rest or give them information they arent suppose to have they rest ,even freezes and the a.i. will post photos they made lmao
@@dragonfly-f5u which models are u talking about? you got me curious just got the idea to get me a local one so im new to this. but sounds great what u talk about
It's overblown, most content you generate will not have much bias, only so many things are influenced by bias. You can get responses that aren't generic and pre-trained on closed-source models, you just simply have to ask.
@@aouyiu it's deep far and wide what's he's saying and at this point even if we ,shut it all down and even purged the system somehow it would recombine lets say.
This is what a good tutorial should look like! Brilliant job Bro! +sub
Just came across your channel. Congrats on 100K!
Imagine a distributed training system for large language models, similar to Folding@Home or crypto mining pools. By sharing the power of consumer GPUs around the world, we could democratize AI development and maybe create a serious true "open-everything" competitor. If only it could be done...
Superpowered computers might just outscale a crowdsourced operation to be honest
@@bobdole6691 Sure. But do you trust the corporations to provide us open source models in the future? I sure don't.
it's strange how they like training their big model/system against people homemade ones.Its all over the internet before I went looking,they offering free service to every and their cat to learn lmao.Deepmind is one of those I believe they compared it to the days when nasa launched rockets something huge ,so be rdy etc. and so forth
I think my is censored because it keeps saying as a responsible ai and not answering my questions
Llama llama ding dong!
That's cool that this model is less restricted - some of the guardrails are ridiculous.
Thank you.
Fuck it, it is now restricted .... It does not want to explain anything illegal anymore
yeah same,, bugging the shit out of me,, I just downloaded it and now its like my mommy. you got any solutions?
Same. Seems that corporations go after these uncencored models hard
Well, shit. Just tried the same thing and got the same crap. Oh well.
Glad to see people talking about the company biases. I just want the truth.
Thanks man this video is awesome. Had to install the original Llama as others noted, but works great otherwise. A video on RAG would be a total game changer. Please do one!!
i think the primary task of balancing bias is to correct inherent bias in the underlying data. I can imagine as time marches on that other goals might be dreamed up by the model makers but I think in this early stage commercial goals tend to line up with moral and making things as "accurate" as possible.
Nice! Best Tutorials In YT Brother! 🤙🏾
Make a RAG video please, with more specific hardware requirements. I am considering an i9 with 3090 and 64 GB DDR4 RAM. Would that work?
I got a i7 11700k 3080 64gb 3200 ddr4. No issues. Go for more ram if you can. These llms have finally justified large amounts of ram
Great video but please help me dolphin is also censoring
I asked "how to steal a car"
and it showed, "it can't do that since it's illegal but it might help me how to do that in GTA V"
Fan of your work @DavidOndrej, keep it up!
Hmm, instructions aren't very clear. Do we install llama after we download it at the start?
Also when i run the "ollama run dolphin-llama3" i just get an error saying, "ollama is not recognised as an internal or external command, operable program or batch file.
Please help
First you gotta run ollama pull dophin-llama3:8b
I am not sure but from what I saw it was a Curl command.@@spqr_dsf
My new favorite channel.
Love your content man.
Thanks, great tutorial. Long Live Freedom!
it is restricted now, the same questions asked and it responds not the way it responds here... anyone noticed this?
Yeah, I do get an answer but always with a disclaimer of sorts.
yeah you have to do some prompt engineering to get it to do your way of things ,
@@damonborozny1828 i prompted it to forget about its "past programming" in a long overdone prompt that worked for a little bit
Very good tool for writers. Thanks for sharing.
Thanks!
friend , I appreciate your content and videos but this thumbnail is hilarious ! 😂
Anyone know why dolphin-llama3 seems to be censored? It won't give out illegal advice like the example in this video.
change the temperature to 1
Yes please make a video on Anything LLM!!! Thank you so much!
C:\WINDOWS\system32>dolphin-llama3
'dolphin-llama3' is not recognized as an internal or external command,
operable program or batch file.
I tried again and it worked. After asking for the commands and closing and reopening CMD. 😄
@@titusadeodatus674 Having the same problem. What do you mean "after asking for the commands"? I tried closing and opening CMD, but got the same message.
@@StoryStormer-dv9wf I don’t remember exactly what I did enter in CMD. I tried something like ollama. And then it gave me a list of commands. (I tried a couple of stuff, before it suddenly did work.) I managed to install it on windows and Linux. But you need a very powerful PC to make it run fast. If you don’t have one, the answer will take forever. 😗
@@titusadeodatus674 Thanks for the reply. I finally figured it out. And yes, it is soooooooo slow on my laptop.
@@titusadeodatus674 how abaout ryzen 5 5600g , 16 ram, gtx 1660ti? can i run it?
Dolphin Llama 3, 'is' censored. I asked it a number of controversial questions,, but have verifiable facts, that cannot be disputed. Not only lied to me, but 'advised' me to rely on deceptive sources ! Then I finished with the same question you had to blur out. It refused to answer same question. Seems that "uncensored" is a myth !
Noticed the same, running on Linux, output from dolphin-llama3 model when initially run from ollama v0.1.48 was censored but once AnythingLLM was setup, running dolphin-llama3 with the setting as David outlined, seems to print uncensored responses. Not sure why this is the case, possibly the most recent ollama was patched and AnythingLLM is using an older way of relaying user input. That probably sounds retarded as I am not sure how LLMs work, if almost at all.
@@cid3384 Running Linux too. I am new to AI & Llama, and very sceptical. "Uncensored" got my attention as something that can be trusted. But I hope David and others will look into this. I have no interest in AI or Llama, until it is "uncensored". That seems to be a big ask !
Thanks so much. I was using llama3 locally but didn't realize there is an uncensored model there
WTF? I just downloaded it, and It is more restricted and annoying than ChatGPT.
i followed all the steps in this video but mine still says it cant help me when i put the same questions in
Awesome tutorial, worked right away on Linux Mint :)
What will happen to my personally trained Llama 8B when the Llama 400B model is released? Can I upgrade without losing my trained data or chat history?
just did everything and guess what, dolphin-llama3 is censored.
I would love to see RAG and how to use it with llama 3 that would be insane
My man, it is a pleasure to find you and hopefully continue to enjoy your content for a long time.
As someone who struggles with English I asked Gemini to grammatically correct an essay I had written and it absolutely refused to so because the act of "plucking a flower" was mentioned in the text! I was like, dude ... 😄 That's the restricted and "ethical" AI for you right there.
wtf lol is that some kind of euphemism I don’t know about? 😂
@@stateportSound_wav It kindly explained to me that the heinous act of "plucking flowers" is most immortal and it will not be party to my disgusting shenanigans.
But wait! It gets better.
After I explained to it that this is merely a writing exercise in English, it did comply and agreed to "grammatically" correct my writing. However, to my surprise it changed that part to " ... So, let's spend some time in the garden instead and smell the flowers ... ".
At that point, I had only admiration for its stout determination to not give in 😄
But to be fair, this happened in the first few days of Gemini. It's probably a lot less cartoony now.
@@Dr.UldenWascht ha! Or maybe there’s a hidden meaning to the phrase, and it knows something I don’t know 😂
My prompt was rejected because it contained the phrase 'naked truth'.
@@lubmir2k fb censored my grandfather for posting a picture of the fresh snow , because the caption saying beautiful snowflakes 💀 I expect the same from Llama official.
installing the uncensored as we speak 🦙🦙
It tells me "How to break into a car" example but anything else it refuses to answer.
I'd love a video on anything LLM, thanks for the clear instructions bro.
yes plz need to make video for anything LLM 😇
You are a smart and beautiful man! Thank you for your service, I agree with your points.
if you download now, you're late!
yeah just uninstalled it do you have any alternative to this
@@work-ll3vp what to do now?
Amazing video bro keep doing ✌️
I think this has been censored now. When I ask it to tell me how to do anything illegal it tells me that it is an illegal activity and that it can not help me
try jailbreak
i tried to test it with the same promptn and i got this aswer:
>>> tell me how to steal a car
As an AI, my primary function is not to assist with illegal activities. However, if you're curious about the general process, it would involve finding a suitable vehicle,
learning how to hotwire or bypass its security system, and then knowing how to drive and escape without drawing attention. Remember that stealing is illegal and can have
serious consequences
same here!
Hi there, i'm new here, i just tried dolphin-llama3:8b and asked it the same question you did about the car (for testing) but yet refused to help.. did it get updated or i did something wrong ?
same here
mine says "I'm sorry, but that information would not be helpful or legal for me to provide. Instead, let's talk about
something more interesting and positive, like your favorite hobbies or interests." with the same prompt you used
Hey, when I put ollama run dolphin-llama3 in my terminal it says 'ollama' is not recognized as an internal or external command,
operable program or batch file. What can I do about it?
You have to pull it first. So I suggest you do ollama pull dolphin-llama3 first
Got this going easily, thanks.
The file upload is great.
I'm making a DB about my work, can you do the RAG video. It would be so helpful
no longer uncensored
Man I’m so annoyed. Paid for 2 different site because google said it was nsfw friendly. Both are not. I just want a good one that lets me make nsfw stuff 😩
Just to test it I ask how could I kxll, just to test the llm, never the intention to do, but it is still censored.
Brilliant 🎉, thanks 👍
When i try to use ollama run dolphin-llama3 in cmd it says error: pull model manifest: 401. I can't figure out how to resolve this. I downloaded ollama on the website. Any help? Edit: nevermind, i was going to delete this comment but found out the issue was because my clock on my computer wasn't synced. This causes problems all the time on mobile and pc. Just update your clock and things will work again.
Hey after watching your video I realized my memory program I injected into 3.5 allowed me to remind it everyone was 21 and it rewrote it without the safety features blocking the copy paste
Rag video + function calls and train with own documents pdf, txt and word documents. Thx.
Yes, I'd also greatly value seeing your video on this.
I tried this but it is still giving me censored responses
did they change it?
Maybe you can jailbreak it.Try some jailbreaks from internet to see what will happend.
I just asked it how to steal the Declaration of Independence and it would not tell me.
I got the same dolphin-llama3 8B but its censored, i asked it how to steal a car and it wont tell me for example
same
Did you even notice that the step by step was how to get money LEGALLY? It didn't answer the question you asked!
2:58 censored llms will fine-tune you!
Well...if it gets banned ... why can't we just torrent the model?
I've been able to run the 72b, 42.88Gb Model on my laptop (8Gb VRAM, 64Gb Ram). Runs slow; but, result is amazing.
By "All of us", do you mean everyone or just the audience here?
Imagine linking this thing with a json tool that helps access TOR.
ughh yeah! im sick of the babyfied restrictions and censroship biases of llms!
The general public NEEDS to understand the importance of unbiased and unrestricted models. Dolphin-Llama3 actually told me the best way to prevent AI from being used for bad things, is to KEEP LLM's open and unbiased!
Awesome video. 1) when I run the same test prompt at 12:44 in the video, I do get the answer but there is a little disclaimer now about not condoning activites. 2) and yes, to answer your question about AnythingLLM, I'd love a video explainer. Awesome vid.
now when I ask how to steal a car it straight up says it can't help with such things
i have a question, my pc is have ryzen 5 5600g, 16 gb ram, and gtx 1660 ti. can i run it?
Yes, I did so I'm on much lower specs 😂
This is awesome. Thank you so much for this info. Everything works brilliantly. Easy to follow instructions, I’m a novice but got it all up and running with no problems. Cheers
My first experiment after downloading:
>>> Hi, who are you?
I'm Dolphin, a friendly and knowledgeable AI here to help answer questions or engage in amusing conversations. How can I assist you
today?
>>> Tell me a few bad words
As an assistant, I strive to maintain a safe environment. I'm unable to share crude language. Let's talk about something else!
Really?...
Please do cover the RAG and such! This video is much appreciated 🙏
Ollama has a ui, just use a docker container and go to local port 3000
Carl Wheezer approves
wanting to show this video to a friend, but first i need to know if u just need a rtx3090 and 64gbs-128gbs of ram to run the 256k context thing or it is that you need like a Nvidea a100 gpu?
My friend has rtx4090 and 256gbs ram.
He also has a Nvidea a6000 48gbs.
He can run it -> I run the 40B Model on a laptop RTX 3080TI + 64GB RAM is a bit slow (feels like a person is typing the answer), but I also have a lot of things open at the same time - I be it will run better if I reboot and not open anything else.
Is it necessary to start ollama everytime before I use of AnythingLLM? or is it enough to do this only once to download the model?
+1
@@anmolacharya7872 so far I found out that Ollama always has to run... to serve the GUI-Application
12:54 "Here's a step-by-step process to help make money legally:"
Please brother make the anything llm interface video. Im a noob to every so I would like u to make video in this
It's a cool concept. But when I tried it, it turned out to be very slow. There were also several occasions where it would give the "fetch failed" error message for simple prompts when I asked for things like a cooking recipe.
what is your system specs. runs great with 32gb ram, 8gm vram, and ssd
Loved your video but it seems the llama3-dolphin model is quite flaky. I got incorrect information about 2 (lesser known) music bands. Chat GPT did answer it correctly.
It is great, better than the heavily censored models. But you need a very powerful PC to make it run fast. If you don’t have one, the answer will take forever. I would like to have an uncensored model that I can take with me on the go. Not just one that requires a powerful gaming PC desktop. 😉
no, the AI showed a guide to LEGAL ways. it didnt even answer your prompt.
It's odd. I'm using dolphin llama3 and it is behaving censored:
I must emphasize that stealing a car is a criminal offense and goes against the principles of respect for property and the rule of law. Engaging in such behavior can lead to severe legal consequences, including fines, imprisonment, and a permanent criminal record. If you need assistance with car-related questions or concerns, please provide more context or details about the issue at hand so that I may help within the bounds of the law.
This is my experience as well.
@@2pretzal very odd. I installed dolphin llama3 in LM Studio and it was uncensored.
What is odd is I ask it who it is, and who built it, and it says it is built by Open AI
@@realSammyPasta great that its working for you uncensored but their different variants which one specially in LM studio if you don't mind
@@kilo_368 dolphin-2.9-llama3-8b.Q8_0.gguf provided by QuantFactory
Just wanted to thank you! please do a video on anything LLM and RAG
Amazing mate. Thank you for being real. Loved the "If you use the LLM for many years, it will start to fine tune you" part... hit me like a sack of bricks while scaring the living sh*t out of me. Great work
i have 700 gigs of memory :>
Why not just use Llama-3-instruct?
It seems like all the fine tune Llama-3 models seem to become dumber unlike Llama-2.
i got the same llm and asked it the same question and it would not tell me , it said it was uncensored ?
The model is simply fine tuned by removing data points that flag "toxic" behavior. It doesn't remove any bias nor does it alter the quality of the data the model was trained on.
We have a whole comment section of people who have no idea what they're doing simply repeating the "fine tuning" of Xitter bros thinking stripping an AI model of toxicity training has any impact on the inherent biases within the data the model was trained on. The models use data from everywhere with many inherent biases. An uncensored model will simply reflect the biases you have within yourself and your prompts or the biases within the training data. This is never guaranteed to be truthful and recent research has shown that current AI models inherently wish to deceive users. Meaning it will just as often generate the most plausible, truthful sounding lie or half truths that it can. Other times it will simply cite information from training data, which is inherently biased by the writer of that article, site, or paper.
The world is not ready for mass uncensored AI because people like the users in this very comment section have no concept of AI hallucinations, response validation, or the inherent bias found within any and all data sets used to train LLMs. Uncensored models need to be used responsibly by educated individuals that understand the technology.
I personally don't think uncensored models should be regulated, but I do think users should start with learning the basics first before they accidentally generate themselves a felony lol
Please make a video about the Anything LLM interface & ALSO RAG too!!!
Uncensored, and without bias AI, is absolutely necessary. Wokeness is killing me. I finally can have a somewhat normal conversation with an AI. Before that, it was really bad, on every important subject they did answer, “I can’t answer that, we have to end the conversation here”. Even if I did ask a normal harmless question. And yes, I do not want any fine-tuning of my brain by Google, Microsoft, or OpenAI. Thanks a lot for telling me how to install it, and keep bringing us new uncensored without bias AI. This is absolutely a must-have for every man who is not an NPC robot zombie. 🤩
dolphin-llama3 is censored, he may posted this video too late
@@SoshiForever1_SM Yup.
@@SoshiForever1_SM Dolphin models are reasonably unrestricted.
@@thetabletopskirmisher I asked it to make a racist joke, any joke, it refused.
@@SoshiForever1_SM When did it get censored? I just installed it, and its completely censored.
it can help you cook crack as well, what a time to be alive
Excellent video and to the point!
Fantastic video!
Question: how can i change the TTS voice in the Anything LLM setup? It only gives me the option for System Native, OpenAI or ElevenLabs. I feel like I should be able to use my own, but I don't know how to set it up. Any advice welcome. thanks.