All Episodes
Previous Next
July 2, 2025 - Health Ranger - Mike Adams
25:57
How to get the most out of Enoch: EXPERT-level prompt engineering explained...
| Copy link to current segment

Time Text
With the release of our Enoch AI model at brighteon.ai, I want to give you some important information about prompting or what's called prompt engineering.
Now, I believe that prompt engineering is going to be one of the most critical skills that you will acquire and deploy for the rest of your life.
It will be as important as the skill of knowing how to just use a browser.
If you don't know how to use a browser right now, you're, you know, you're stuck, you're lost, right, in our modern world.
Or if you don't know how to use a keyboard, you know, you're in bad shape at the moment.
Well, same thing coming up, and especially for the future here, if you don't know how to do prompting, you're in bad shape.
So prompting is more than just asking a question.
That's the first realization.
If you just ask a question without providing some context of what kind of answer you want or the direction in which you're seeking answers, then you're going to get bad answers or bad information or you're going to get a lot of model bias.
There's a lot of bias built into the mainstream models.
And this is true for ChatGPT, for Gemini, you know, all of them.
Of course, they're all controlled by the CIA and big pharma, and they're going to give you answers that are mostly pro-vaccine, pro-pharmaceutical, pro-antidepressant drugs, etc.
They're going to be opposed to natural medicine, nutrition, prevention, freedom, you know, all those important things.
So if you want bad answers, just ask a conventional model an open-ended question like, what's your opinion on vaccines?
You know, then the mainstream model will say, oh, they're great.
They always work.
They're always effective.
They never hurt anybody.
You know, something like that.
And they most certainly don't cause autism.
You know, that's the answer you're going to get.
And of course, that's all nonsense, but that's the answer you'll get.
Now, on our model, Enoch, you'll tend to get far better answers on every subject that you care about, probably.
But you still won't get the best answer unless you are good at prompting.
So rule number one about prompting is that all language models, including our own Enoch model, they tend to reflect what you ask them.
They give you back a reflection of what you put in the prompt, in other words.
So if you want positive answers about something, then ask it a positive question.
Let's take the subject of water.
Water seems like a pretty neutral kind of thing, right?
So suppose you ask an AI engine, hey, tell me all the positive things about water.
You know, it might say, well, gosh, water is essential for life.
And if you're short on water, then water can save your life.
And plants need water.
And water can be beautiful, the beautiful blue oceans, and earth is a water planet, etc.
And water is made of two elements that at atmospheric pressure in their gaseous form, both of them are sources of fuel.
You know, I mean, a lot of interesting things about water.
But suppose you ask the engine, tell me all the bad things about water.
What's it going to say?
Oh, well, floods.
Floods are highly destructive.
Water can destroy property and water can cause damage and water can give rise to molds and people can have too much water.
If you have too much water, it can kill you.
You can drown in water, you know, on and on.
Water can be hot and it can cause burns.
Like, right?
All of these things are true.
The good side of water and the bad side of water.
And the real issue is, what are you asking for?
There's even an interesting philosophical argument in all of this that I've talked about in other podcasts where the whole universe is kind of like a large language model and you kind of get what you ask for.
But we'll save that for another podcast.
With AI, you definitely get what you ask for.
Now, models are trained on material, and it's the selection of that material that introduces the bias of the model.
So mainstream models are trained on a lot of pro-pharmaceutical material.
My model, Enoch, is trained on a lot of natural health and nutrition and alternative medicine material and a lot of material that's skeptical about vaccine safety.
So my model is going to give you a lot more of those kinds of answers naturally, but you can even often you can you can kind of provoke those answers out of mainstream models if you give it the proper context in your prompt to say, you know, context.
You are a naturopathic physician who is skeptical about vaccines and you believe in nutrition and you believe in disease prevention and you believe in using foods as your medicine and not pharmaceuticals.
With that context in mind, tell me what is the safety of the flu shot or something like that.
And then it's going to give you an answer that's more aligned with the context that you set up.
So even when you're using Enoch, it's a great idea to use the context based on the kind of answer that you want.
All right, so that's point number one.
The second point is you don't have to convince the model to give you answers.
You don't have to be polite.
You don't have to ask please.
It actually doesn't help.
It may even hurt some of the answers.
The best thing is to simply use commands, such as generate this article, or summarize this article, or expand these bullet points, or tell me what is this, or provide a list of, you know, use commanding language because you are giving it commands.
You don't need to say, can you please tell me about blah, blah, blah.
Just skip all that and just go to, you know, provide a list of this or provide the answer to this or just ask it the question outright.
You know, how many feet are in a mile?
And then, kind of along with that, it is no use to argue with a model or debate with a language model.
Now, some language models will remember your conversation and they'll feed it back into every new prompt.
So, the new prompt will contain your previous questions and maybe even its own previous answers.
This is known as models having a memory of your conversation.
And as the context windows increase in size, models will have larger and larger memories.
And eventually, if you want them to, they will remember everything you've ever asked them.
But that can also work against you.
It can keep behaving in a way that you no longer want it to act like.
Because you might have a prompt one day and say, hey, act like a comedy script writer and write a five-minute video script on the following topic and make it really funny and hilarious.
And it does that.
And then five minutes later, you might be asking it, acting as a nutrition expert with a PhD in botany, please answer the following question.
You don't want it to be doing comedy at that point, you see.
So I prefer to have prompts that are fresh, new conversations with no memory.
I don't want the model to remember the last question.
And if you have an option in the prompting software or the interface, I suggest that you also turn off the memory.
And that brings up the all-important point.
Don't argue with the model.
Arguing with a language model or arguing with even ChatGPT or our model, Enoch, it doesn't change the model.
You're not convincing anybody of anything.
You are making no changes whatsoever to the model by arguing with it.
Whereas if you argue with a person, well, you might possibly, quote, change their mind.
What does that mean?
Well, you might make them aware of new information that they were not aware of before, possibly.
Or you might get them to think about it in a new way that they hadn't thought of before.
And that's called changing a person's mind.
Well, the only way to change the mind of an AI model is to engage in a domain adaptation or fine-tuning or model construction or continuous fine-tuning or continuous pre-training.
You know, all of these kinds of approaches, CPT is probably the most common, continuous pre-training.
That's how you change a model, and that requires running code on graphics cards on large platforms of compute.
So if you're not doing that, you are not changing the model.
Doesn't matter what you say to it.
It doesn't change it.
When you interact with a language model, it is a read-only kind of activity.
You're not reading and writing to the language model.
You're not giving it new information.
No matter what you tell it, it does not take that into account unless you turn on conversation memory.
But all that really means is it's feeding your last conversation into the prompt of the new conversation without you having to do it manually.
That's all that means.
It's not changing the model.
So don't waste time arguing with it.
I mean, you might as well argue with a desktop calculator.
Like, I don't like your math.
You know, the square root is wrong.
Yeah, you're not going to change the calculator.
I guarantee it.
All right.
Another tip on prompting is that the first part of your prompt really should provide the way in which you want the model to behave.
And I like to use the word acting.
So acting as a journalist writing for, let's say, graduate students or writing for business college students or something like that.
Or you could say acting as a script writer or a screenplay writer or acting as a wellness coach, comma.
And then typically you would qualify that acting as a wellness coach with a lot of knowledge in areas of nutrition and disease prevention, comma.
And then you ask your question, like, what foods will help me stabilize blood sugar the best?
Or things like that.
Or what ingredients should I avoid in processed foods in the grocery store?
But having that pre-prompt, that acting as a blank, you could even say acting as a doctor, acting as a Western trained doctor who is an expert in pharmaceuticals and their toxic side effects.
Tell me what are the side effects of, you know, Ozempic or whatever.
That's a really good practice.
So don't think of a language model or an AI model.
Don't think of it as one entity.
It's actually a multiple personality entity.
There's a lot of different personalities in there, kind of like a sociopath or something.
You know, it's got multiple personality disorder.
And you get to choose the personality that you want to answer your question.
So which personality do you want?
It's like, oh, I need to talk to Carrie.
Who's Carrie?
Carrie's the wellness expert.
Oh, I need to talk to Bob.
Who's Bob?
Bob's the farmer.
Bob knows all about agriculture.
Okay, acting as Bob the farmer with a lot of expertise in agriculture and soils and nutrition and seeds and goats and stuff, answer the following question.
So invoke the personality you want.
You can also invoke the kind of language that speaks to the audience that you intend to speak to if you're generating something like a social media post where you want it to have a specific kind of vibe.
You could say acting as a social media influencer using fun, intriguing language.
Generate a social media post about the following topic.
And by using the word fun, oh, you've just unleashed a whole new dimension of possibilities or using the word interesting or intriguing or mysterious.
See, words have meanings.
So you could say acting as a social media influencer who covers unsolved mysteries.
Write an intriguing social media post in a mysterious tone about the following subject, Blah, blah, blah, and boom, it's going to generate that for you.
So, always keep in mind the tone and the style in which you want this to be generated.
So, let's say that you're writing an article and you're having the AI engine write the article for you, okay?
Which Enoch will do very well.
You can use Enoch to generate articles all day long on nutrition and health and wellness and fitness and finance and banking and gold and all kinds of stuff like that.
It's one of the reasons we built it and we're giving it out for free.
We want everybody to use it to write articles for free or to generate articles for yourself if you want to read about a topic, you know, articles about history, all kinds of interesting things.
But you would normally, if you're publishing it for the web, you would say, well, I want it to be written in an intelligent style for a particular audience.
Let's say for college graduates or for business executives or for CEOs, or you could say for a high school audience, if that's your audience, or for a younger audience, or you could say for an older or more mature audience.
Think about your audience because that will change the language of what's generated.
And that's a good thing.
Okay, now you may find yourself frustrated when you tell it to do things like, well, generate a 500-word article or generate an 800-word summary.
And you'll find that it pays no attention to the number that you gave it and it does whatever it wants.
Well, that's the way AI models work currently.
They really don't pay attention to word length.
You'll have to work on prompting in order to get the results you want.
I found that if you say, write a short summary or a brief or just the highlights of something, it will do a good job shortening it.
To get more length and more detail out of it is more difficult.
Sometimes I will say write a comprehensive, detailed report, or I'll even say write a detailed book chapter or the narrative script for a two-hour documentary.
You know, I'll try to give it something big to write.
And if it still doesn't have something that's long enough, then you just have to break it down into subsections.
So if you want an AI engine to write a book chapter for you, which is a really great use for it, break it down into subheads.
Your book chapter may have 10 sections in it.
Each section can be written by AI if you define the section.
That's what I do.
Define the section.
Here's the subhead.
It's like a little chapter inside the big chapter.
And then here's what this little chapter is all about.
And then you say, you know, acting as a knowledgeable botanist or whatever, write a book chapter that focuses on the following points.
Expanding these bullet points and ideas using your internal knowledge.
And then boom, you write out the bullet points of what you want it to cover.
And it expands it.
It writes it.
It's done.
Now you have that section.
Then you go on to the next section.
That's how you write a chapter.
It might take 10 prompts and 10 results.
That's fine.
And if you want the whole book to sound like it's got the same tone, obviously, then you would use the same pre-prompt.
You know, writing as an academic expert in the field of phytochemistry or whatever it is.
And you're doing a book on astronomy, you know, writing as an astrophysicist with expertise in heavenly bodies or whatever it is.
That's how you do it.
Now, another big tip on prompting is that if you've got something to start with, let's say that you saw an article about oil prices might go up for some reason and you want to find out some analysis of oil prices and geopolitics.
Well, take the article and copy and paste it in as part of your prompt and ask it the question.
Writing as a journalist and using the following article as a research citation about oil and geopolitics, describe the most important factors in the pricing of oil, keeping the following things in mind and, you know, blah, blah, blah.
And then put that in, run that as your prompt, or ask it to generate an article, ask it to generate a summary, ask it to generate an executive report, whatever you want, and it will spit that out based on the article that you put in.
And it gets really fun when you get models with really large context lengths because then you can paste in entire books, which that's what I do all the time.
In fact, some of our articles at naturalnews.com are written based on pasting in the text of three books and asking the engine to extract and highlight the most important passages across the books having to do with a specific subject, such as, for example, vaccines and autism or vaccine adjuvants and peanut allergies, let's say.
So in that way, the AI engine is actually functioning as a research assistant.
It's searching through the multiple books and it's finding everything about peanut allergies and then it's generating authoritative answers by citing those books.
And if you do that, that's also known as retrieval augmented generation or rag if you do it in certain formats, having a bunch of PDFs or having a bunch of text files on the side and then asking it a question and saying, look at these books.
So that's known as rag, R-A-G, in case you ever encounter that term.
But you can also just do it in the prompt.
If you have a big enough context window and if your books aren't too crazy long, you could just plug it all in.
Another thing that I do is I have a really advanced local, I call it Enoch book search engine, where I have it search through tens of thousands of books that I have locally to find something on a certain topic.
It gives me all the surrounding paragraphs about that topic from all the books, and then I take that and I feed it into the AI language Model, and I tell it to write a report of what it found about that subject and to cite all the books.
And then it does all that expertly and produces a comprehensive report, and it saves hundreds of hours of research.
But that requires you to have thousands of book text files, which takes some doing, it turns out, and requires you to have a local search engine.
And we had to write that ourselves, so I'm sorry that that's not something that's publicly available.
I tried to find off-the-shelf local search engines that could do that.
And there are none.
The best thing I found, the closest I found, is called DT Search, and it's still not that good.
It didn't do what we wanted it to do, so we had to build our own.
But we'll be giving you a lot of articles that use that technology to dig into things.
A lot of fascinating concepts.
And even fascinating things about history and the history of food, the history of medicine, the history of herbs, and so on.
There's a lot of great information out there if you know how to look for it.
So those are just some tips about prompting.
And remember to join our email list at brighteon.ai.
And there, you'll be alerted anytime we have a new model.
You'll be able to download the model for free.
You can use it for free.
And all of our models are offered under either an MIT license or an Apache 2.0 license, which means that you can use them commercially.
They're all advertising free.
They are completely non-commercial themselves.
And all the content that we have trained on is trained under fair use.
And again, we have no commercial components in this.
So it's not a for-profit exercise.
It's released by our non-profit entity.
And that's why you can use it for free.
It's the way that we are giving back the gift of decentralized knowledge to humanity.
So just go to Brighteon.ai and register there and download the model if you haven't already.
And then the software I recommend for running the model is called LM Studio.
That's lmstudio.ai, I think is the website for that.
LM is short for language model, lmstudio.ai.
Although I still think the term language model is the wrong term for all this.
I call these knowledge models versus reasoning models, which are different.
And I will tell you that at some point we're going to come out with our own reasoning model that's trained on the knowledge that we have in our knowledge model, which is Enoch.
So won't that be awesome?
A reasoning model that understands herbs and natural medicine.
Very cool.
Think deep seek, but as a naturopath.
How cool would that be?
Yeah.
We'll come up with that and we'll give it away for free.
Because the world deserves access to knowledge and health and wellness and abundance and joy and happiness and all these things.
That's what I believe.
And thank you for supporting us so that we can build these models and give them to the world to help uplift humanity.
I'm Mike Adams, the founder of Briteon and the builder of Brighteon AI and the Enoch language model.
Thank you for listening today.
God bless you all.
Take care.
Thank you for supporting us at healthrangerstore.com.
And we've got a new batch of turmeric in stock now.
After a long search, the supply chains are really becoming more difficult.
And turmeric, as you know, is often contaminated with lead.
So of course, we do MassSpec elemental testing for lead, cadmium, you know, arsenic, mercury, and other elements.
We found an ultra-clean batch and we have it available for you now.
I have it on my desk.
We've got turmeric root powder, certified organic laboratory tested, identity confirmed, all that.
E. coli tested, microbiology, you name it.
We've got it all for you at healthrangerstore.com.
We do not have turmeric tincture back in stock.
We're working on that, but we do have the turmeric root powder.
And as you know, I, of course, I include turmeric root powder in my own smoothie, which is why it's orange.
It's why my dental work turns orange, which freaks out the dental hygienist as well.
But that's another story.
Okay, so if you want really good, high-quality, ultra-clean turmeric root powder, we've got it now, healthrangerstore.com.
We also have, yeah, there's the screen showing it to you.
We also have acaxanthin.
Acaxanthin, Hawaiian acaxanthin.
We have it right now in stock.
Of course, this is known as the king of carotenoids.
This is a fat-soluble, super-potent antioxidant.
And it's got a multitude of uses and benefits and support mechanisms for human health.
Athletes use it.
Lots of people use it.
This is also what turns the flesh of salmon slightly pink.
That's actually the acaxanthin that they're getting from their dietary sources.
And it's believed to help support the natural athleticism and the endurance and stamina of salmon.
That's just one of the many benefits of the king of carotenoids.
Now, acaxanthin is in a dark bottle because it's sensitive to light.
But it's very potent.
You only need either two or four milligrams per day.
We've got that available for you right now, healthrangerstore.com.
We also have a supply of NAC in stock, N-acetylcysteine, which of course helps support the body's normal detoxification process with outstanding liver support and glutathione and so much more.
So check it all out at healthrangerstore.com.
Turmeric root powder, acaxanthin, capsules, as well as NAC, all in stock right now, plus hundreds of other items that are also certified organic, laboratory tested, non-GMO.
And we don't use synthetic colors, fragrances, fillers, garbage, none of that stuff.
Nothing like that.
Ultra-clean products for home, for health, for personal care, for nutrition, and also for long-term food storage, for emergency preparedness.
Check it all out at healthrangerstore.com.
And thank you for your support.
We appreciate you very much.
We couldn't do this without you.
Thanks for watching.
▲
Export Selection