hpr-knowledge-base/hpr_transcripts/hpr4101.txt

Episode: 4101
Title: HPR4101: A I O M G
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr4101/hpr4101.mp3
Transcribed: 2025-10-25 19:38:23

---

This is Hacker Public Radio Episode 4101 for Monday the 22nd of April 2024.
Today's show is entitled AIOMG.
It is hosted by Operator and is about 28 minutes long.
It carries an explicit flag.
The summary is, some other guy on the internet talks about AI.
Hello everyone and welcome to another episode of Hacker Public Radio with your Operator.
So today I'm going to be talking about AI, right?
I'm going to talk about stable diffusion, my experience with it, and silly tavern in AI
text generation.
So buckle in, it's going to be quick, fast, and I'm going to go ramble through my notes.
Also work AI resume generation, G's prompt engineering, I'm going to share all that.
So anything you're going to do with AI and in my current state with AI.
So we'll first start with the stable diffusion notes.
I have a master docs that I am putting all my notes for AI in because it's all disparate
all over the place.
I do have some silly tavern stuff to add, but it's pretty click and go.
So first thing I'll kind of talk about is just I'm just going to go order in the documentation
here.
I don't have any examples.
So anyways, so I want to say back in September, I started playing with stable diffusion.
I have a cheap like a gig card or something like that, a gig card.
So it's pretty slow, can't do a whole lot, can't train models very easily.
That might have all changed, but this was back in two years ago, a year ago.
So anyways, stable diffusion, it's pretty easy to set up.
Jamboree has a one click installer for it.
So I'll do Jamboree.
Have I done one on Jamboree yet?
If I haven't, I need to do a one on Jamboree.
So Jamboree links, check if you've done Jamboree HPR.
It's for AI notes, doc, cool.
So resume update tips I have in the document.
This guy is like, you know, your SEO search engine optimization, dude, bro.
But he seems to make sense.
I have resume notes that are hard that he has on his site, but you have to sign up and
click and do an email and click another thing and log into this thing and share information
here and do this.
Then you get a PDF.
So I have that extensive resume notes in there.
I have a ripped version of cyber SN, which is for cyber security jobs.
Basically that's a questionnaire, a bunch of questionnaires they ask you about your
proficiency in each area of cyber security.
And that was in 2022, so I should probably run that script again.
But there's a core in that sub directory.
There is anything in here, anything in the no before or no more cyber SN.
So anyways, perfect chat GPT prompts, basically there's building blocks for them persona.
This guy's, of course, his view of how it works and it makes sense.
You've got persona context task, format, temp bar, and tone.
And some of these are not required for it to do what you want to do.
Maybe I think format and maybe tone is not needed, but anyways, I kind of break it down.
I show a screenshot of his video.
I show an example.
I show another example with my own context.
And then I have like example responses on onboarding scene guidelines, hearing incident
response management.
I don't even know what's called Sims.
So what else do I have?
There's AI agents, Professor Synapse, you can Google that, p-r-o-f, Synapse, no spaces.
So it's a sort of automated, it's a one-click prompt agent.
And if you have the full blown version of chat GPT, it will use the following things,
plugins, whatever it's called.
It will use relevant tools, vision to analyze images, web browsing, advanced data analytics
or Dolly.
And those are like the tools or plugins that it calls.
I don't know if that actually works as I'll have the whatever, I use kaji.com.
I'm not going to talk about kaji today, if I don't have one in there, let me add one.
Let's see, Kakarolist, kaji.com, they have a beta AI search, that's pretty awesome.
So I'll talk about that at some point, if I haven't already talked about it, don't
think I have, wow, real link, this is all old stuff.
So anyways, we're on Professor Synapse, AI agent, basically it's the basic, I have one
click kaji that has my kaji key, and the entire Professor Synapse prompt, it's a really
long URL.
And people can click that URL and go jump straight into the AI agent.
Now it is missing those tools that chat GPT doesn't have, but it's pretty much, you
can ask it's stupid stuff, and it will ask you questions and it will deploy agents based
on that agent.
So to say, I'm having problems with Azure, you know, Azure buckets and they'll be like,
oh, I'm going to deploy an expert in Azure engineering and scripting, an automation
in PowerShell, whatever.
So it'll deploy agents for each thing and then walk you through that stuff.
So I've had some pretty good experiences.
I was trying to do port forwarding in Windows Supply Simpli Linux, and I went back and
forth and it told me to do the output of a program of Telnet, and I used NMAP instead
and detected the output of NMAP as sufficient enough, and it said, oh, looks like you used
NMAP.
Yep, that port 22 is open.
You know, that looks good.
Run this command, and I ran it, and it's like, oh, looks like you ran that command in
the whole place, we need to do it on the host machine, and it's like, wow, it detected
that I did the wrong thing, basically called me an idiot, and then told me that I should
run it on the host, OS, and not the guest, OS, or whatever.
So that's pretty cool.
Let's see, I think I have, I should probably take this out, actually, Professor Synapse,
yeah, we're not going to, I don't think I'm going to share this, it's going to get
it a, yeah, that's my Kaji account, so I need to like, the history, I guess at some point.
Anyways, oh, but no, it's delete, delete, Kaji, keys, I mean, even if I do get them blown
out by another one, there's like, I love those guys.
So there's some old stuff in here, KajiPT agents, GodMode.space, AutoGPT is also old.
I don't know where they left off at.
There's a billion other ones out there now, fabric from unsupervised learning guy has
done some stuff.
I haven't looked at it.
There's, I mean, there's so many things to look at.
There's a wishful search, there's a couple of other ones I need to look at.
I think just wishful search and the fabric thing, I need to check out at some point.
Check, check how fabric, hey, hey, hey.
But he's kind of making some of his stuff public.
So images you can do, play around with images with tiny pots, the free thing, you don't
need a hoard key.
I do have a hoard key if you are interested in messing around, feel free to send me an email.
I can share my key with you, just don't, you know, abuse it or whatever, but I've got
a stupid amount of kudos.
In the hoard, stable hoard is like a crowdsourced AI token platform.
So you can use hoard on AI art, you can use it on fake news, you can use it on basically
anything that supports hoard as a utility or whatever.
You can use that, looks, we have Mr. bots, real fake news, just triangles, color mind,
I don't know if any of these have random sandwiches.
So I don't know if that any of that has hoard integration, but your AI art and your
AI text or text generation or whatever they call that, predictive text is, a lot of
them have support for, support for, for hoard.
Let's see, bark UI, audio cloning, I think there is one that's just called that for silly
tabern, it's called like boys cloner or something like that, it's very, very standard, I have
not messed with that yet, I would like to mess with it.
I have mess with the limit of labs, it's pretty decent, but it's all AI driven, so you can't
have your own local model.
So hopefully bark will work or this new one that silly tabern uses has integration for.
You know, I've got some example prompts, that's all, it's just a mess, well say in general
there's also that site that has additional models basically, so you have your AI model
and you can attach stuff to it, civati, civati, civati, c-i-v-i-t-a-i.com, and it's basically
a giant repo of all kinds of extra models and basically tweaks to different things.
There's all kinds of crazy stuff, I mean a lot of it's like porn and anime, but a lot
of extra luras which are essentially smaller models, you can attach to a current model
for the lack of a better term and make stuff better, it doesn't really work all that
great, it's kind of, you know, I played around with it for about three months and then I
realized there's no catching up with it, same thing with all the other AI stuff, there's
no way to stay on track, yeah, I think I have like any gig card, I tried some of the local
stuff, the minstral 7b, minstral 7b stuff, I mean it's kind of cool, I just used all
online, whatever, I have a single click Alamo with Jamboree, so you just click Silly Tavern
and then bam, you're up with Silly Tavern, and then automatic level 11, you click that
and you're up with automatic level and then you're going to go, let's see, this is more
some auto dbt stuff, poe.com has a bunch of different, if you want to mess with a bunch
of different models, they're the ones that run that website that's like a stack overflow,
basically, it's a Dutch company, but they're pretty big in that space, all their stuff
is posed run by, all those guys to do, like it's like a, whatever, let's see, I'm
here some personas, so with Kaji at least, you can set up a custom agent, and not really
custom agent, but you can set the personality of that agent with under like, I don't know,
it's like under a thousand characters, so I said be pinky, be pinky from the brain to
TV series, yeah, I find it easiest to understand responses, when the tech has written as it
was spoken from pinky from pinky in the brain, I've tried some more kind of an explicit
word, whatever ones, and of course, I'd be in the middle of a prompt, and it would grab
out, because it would detect that that was bad stuff, and it would not let it through.
Right now, I just called one called just code, and it says provide the code only without comments
or explanations, so let's just straight up, it doesn't explain anything, it just dumps
code, and I look at the code, I'm like, that's not what I want, dump code, so a lot of times
with the pinky in the brain, it would go on and on and on and on and tell you, you know,
here's the field for this, and here's, it helps, but usually, if I'm using the code generator
for code, they have research code and chat for custom, if I'm using code, I just want code,
I don't want to have to like, generally, I don't need all the explanations of all that stuff,
so you can also do like a prompt agent persona examples for yourself, and without actually
having your model up yourself, you can take the model as a brick personality thing,
which we'll go into semantics about if that's, you know, actually legitimate or useful or not,
and then there's an integram type, you can take it integram quiz, so I have a personality
of echo, was it November or Foxtrot, Juliet, and then a tri-type integram of 729,
and you can say, you know, help me refine my resume to be a retirement integram, be sure to
include a concise of the bold points, and write in the style of an MBTI, Myers-Briggs personality,
ENFJ, and the tri-type integram 729. Now, I haven't really noticed a whole lot of difference,
but I thought it was an interesting exercise. I was looking around for something,
and I stumbled across a big repo of like, paid for prompts, which is sounds ridiculous,
someone would pay for a prompt, but most of them were pretty bad,
and this was an example that I saw some guy say is like, oh, over 70,000 different personalities,
and it was like, you know, examples were like, Yoda, talking to the third person,
blah, blah, blah, blah. It's interesting, but you could get the AI to write whatever personality,
whatever personality prompt you want, why would you pay like $3 or $15 for a prompt?
Let's see, the Bro Code 1, personality example, that's not super interesting.
Now, find the easiest to understand responses when it's written as if it were spoken by Dubro.
Please talk and Dubro as much as possible, and refer to me as brah, occasionally ill at your dorm room,
or take about being messy. That one was weird, because I tried to forcefully do it more
to talk about Jake and how crazy or messy he is, and I tried to force it,
and it didn't get as much traction. So I got to sit 10 more minutes.
Training, old stuff, the last minute, say this is all old,
stable diffuses in training, notes, there's other platforms like,
what's it called, something.ai?
That you can train online. If you're going to get into it, just look for YouTube videos
of Google Colab, and you can give them like $50 and you can get enough to do a model
or two training to do online. It's only like 20 pictures or whatever,
so I want of myself and my wife, they're not perfect, but they're decent.
They say don't wear makeup, but I've got to know it's about, you know, training,
AI, faces and stuff, which is probably already already all changed by now,
since whatever. I've got an interview about the AutoDirect,
about the kind of the end of the world, and AI and all that stuff, which I believe is hopefully not true.
And then there's Cambia AI, which is a presentation generator.
So you can say, you know, write me a 10 slide deck wing about how babies are made for
where, you know, middle schoolers and, you know, to the greater, whatever.
I don't know. It'll generate content and it tries.
Let's see. This was supposed to be about silly tavern, and I've got a few more minutes here.
Now, silly tavern, I played around with it a little bit.
I made like a character of my wife, and I made a character of myself, and you know, had some fun with that.
You can do scenarios, and then you can also do, like, author's notes, to get injected to keep the AI on track.
So I don't even pretend to know anything about this. I just played with it.
I had a nephew that was into the dice rolling thing, done just the dragons,
and I gave him some information about this, and then I went down a rabbit hole for a couple of days.
But they have, there's several online ones you can Google online.
There's the one I will give you guys is Cobel Kai, which supports Horde.
The one I like is A-N-A-G-N-A-I dot chat.
That was pretty cool, just by a based one.
But honestly, I would just click the whatever, and start up silly tavern is a node thing.
And it's a lot better, as far as I can tell, just idiot-proof, plug-and-play, it gets things.
You can import easily, there's like, import buttons on the chub.ai where you can get example characters, and import them.
It's pretty cool. It's pretty quick.
Even without having a Horde key, it's pretty speedy.
If you pick guys that are open, or agents, or workers that are like 10 workers or 15 workers.
Because it's very small, and it doesn't take a whole lot to do.
I don't know how those guys do it. I was doing the AI art stuff, so I didn't have any extra cycles or GPU memory to load up anything.
But these guys that are loading up, whatever, 70b, whatever things to take like 24 gig cards for just your text.
So that's kind of the silly tavern stuff. I don't have a lot of input.
You can also do images within it.
So you can essentially fire up a local AI model and a local voice clone of yourself.
And then as the story goes through, you can have the story create an AI art generate art based on the story as it goes through.
And you can also do the AI voice. You can do the voice clone and have it talk as your voice.
So you can go back and forth. So obviously I'd like you to do that.
And you know, do a story with my voice and Kathy's voice clone AI and spit it out to the parents to terrify them.
My kid did one in with a copilot about my boss and he was like the protagonist and my mom called me all concerned.
We'll slip it out and I was like, there's nothing you could do about it. It's coming. It's starting to hear.
Let's see. What else do I want to talk about? Hey, I wise.
I use the, you know, refine refine. I guess I'll talk about Kaji.
Kaji's 350 a year, 250 a year, 250 a year. You get open AI through Popeye Turbo, GPT-4 and GPT-4 Turbo.
It's good. It's pretty awesome. It can be sliders slurry in the mornings on Mondays, whatever.
But you also get inthropic clawed incident one and clawed two. And then you get Miss, Miss Roll, the Miss Roll small and Miss Roll large.
I don't know. And they kind of give you like a little eye chart.
Oh, maybe.
What the difference quality speed cost context window.
I guess clawed two has a bigger, I think clawed two has the biggest in here. Maybe, maybe clawed in GPT-4 Turbo seemed to have the biggest context window.
Anyways, it's like 20, 250 a year. Kaji, it's AI search. It is, let's see, 153, 153 total.
Anyways, I'll go back to 70 per year. They have a 54 per year. It's at 3600 searches per month or something.
Anyways, they have like a family plan, they have a monthly, whatever. So it ends up being around 270 a year or if you want to try it for like 45 to a month, you get 54 bucks.
But it's basically, you can upvote different sites and they have a list of already curated sites that people have upvoted and downvoted.
So like Pinterest is the number one downvoted. I got rid of Wikipedia and a couple of other sites that I know I never want to see results from.
And it has the research into it, which is AI plus search. So you can just type in a question and it will give you the answer without having to see any ads or anything.
They have a API. I might even, you might even be able to make your own API. I don't know if they have any weird MFA or two auth or anything crazy. I think it's just like wide open right now. You can write your own API.
But they have any official API you can ask to get. I think they're only doing it for businesses or whatever. So they have research, which I think is expert, which is fast, fast GPT.
And expert, I think is probably GPT for Wolfram. It's supposed to use Wolfram, which I haven't been able to coax it into using it.
So anyways, research assistant again, and then there's code. Of course, you can pick from GP4 Turbo and GPT for.
And then which I'm interesting, it's not Claude. It's not using Claude. And then I have another one that's the prompt for Professor Stamps, if I want to just kind of generalize about something.
If I know what I'm going to talk about or ask, then I'll use that. But if I'm not quite sure, like, you know, for example, like, how do I keep weeds out of my line?
Like, it'll do things to ask you like, where are you located? Southern hemisphere. What time of the year is it? When should I start? Whatever. And it'll guide you through that instead of having to like, we get a bunch of stuff.
And generally speaking, you know, outside of a few recipes, I haven't had any issues. Now, there's this bus doc, which is basically the same thing as research.
You can put it in URL and it kind of gets blocked.
Third of the time, they don't have any like Selenium scripts or whatever, when they're spidering, I think it's all JavaScript.
They're not using like any anti whatever. So if they hit, if they hit like a known site that's protected by DDoS,
those content distribution networks, they are anti bots, basically. So anti DDoS, like Akamai and CloudFlayer.
If you aim a URL, what those they may or may not work. So you might have to scrape the data and then send it to uploads so you can research and upload.
I've used it to like upload a document, a PDF for a Bluetooth tracker player.
And I asked the uploaded it and then I asked AIA, you know, can I sync more than one device at a time? No. Cool. Done. I didn't have to read anything.
And it parsed it. Code assistant and then just code is the custom AI assistant.
What else? Again, you can send somebody a single link with your key, with your Kaji API key. And they can go straight into doing whatever they want as your account.
So if you want to have a family member.
No, it's on my wife's computer. I should be on the family. I should be on the family plane at this point.
It's also a universal summarizer. It has a what I feel like is a, anyways, it's a Chrome plugin. And you can say, you know, summarize this page.
And then you can like ask questions about it under fast GPT. Through the API, you can have it summarized in different methods.
You can pick which one you want to, what API you want to use. But I use it a lot. People send me a big article. And I'll just poop, summarize it.
One paragraph to three words. And I can usually tell if I want to go back into that and see any more detail.
There's, if you all know of any news automated information security news, AI things, I would be interested in that.
I haven't really looked around a lot. I've stumbled across some and heard podcasts about some, but I don't know of anyone that's like custom tune.
I would like to be able to custom tune my own new seed to watch certain websites, right, and get like an AI feed of whatever the, you know, super the day is.
Also, I wanted to parse, you know, discord slack, matter most and mess it on and put that all into my own custom feed. And that's part of my own agent.
Yeah, agent I'm working on that will help me with communication and a calendar, personal calendar, personal time management, communication.
It's ultimately I want, like I'd say, like, hey, Google or hey, Robert, and it'll pop up on my wife's phone and it'll be in my voice and it'll, she'll be able to ask a questions or schedule meetings or, you know, whatever.
Then it'll help me do more better. Anyways, that's pretty much it. I know I've talked about a lot of different things.
I would say you could get lost in any of them, especially like the, you know, tech generation stuff in the AI photo stuff.
But that's pretty much the high and high and mighty of it. I send a poll links to co workers and stuff that don't know how to do anything.
So I've been on calls and I'll just start talking to Professor Sinefs to walk people through tasks to, you know, things I don't have access to.
You can use Professor Sinefs to walk through literally walk through something. I had Professor Sinefs to walk somebody through setting up something in Azure in an Azure, which I had no access to do.
But it was able to kind of give them 90% of the way there.
So anyways, I think that's pretty much it for like the AI stuff. Again, Jamboree comes with silly tavern and Obama and automatic left living.
I don't really mess with those. I'll do, I'll go jump on tiny pots when I need to like do something funny. Like, you know, if I can see a meme or I want to make a funny thing for somebody, I'll do his head.
It's pretty much it. Again, if you have anything around news feeds, customized news feeds, um, let me know.
That's record an episode. And if you want to co host or be interviewed, feel free to reach out at FRELOAD101 at Yahoo, or you can find me at armacardio.com. Take it easy.
You have been listening to Hacker Public Radio at HackerPublicRadio.org. Today's show was contributed by a HBR listener like yourself.
If you ever thought of recording or cast, click on our contribute link to find out how easy it means.
The hosting for HBR has been kindly provided by an honesthost.com, the internet archive and our syncs.net.
On the Sadois status, today's show is released under Creative Commons Attribution 4.0 International License.