Episode 12
Navigating the AI Agent Landscape: Enhancing Human Productivity
AI agents are poised to revolutionize the way we interact with technology, offering a glimpse into a future where these intelligent systems can autonomously perform tasks, make decisions, and enhance our daily lives. Mike, Mikah, and Artie delve into the various applications of AI agents, comparing them to virtual assistants that not only support but also learn and adapt to user preferences. They explore how these agents can manage everything from personal schedules to complex business processes, ultimately streamlining operations and improving efficiency. The discussion also touches on the evolving landscape of AI, including the integration of advanced features like screen scraping and keyboard emulation, which allow agents to interact with existing software without the need for extensive coding. As they contemplate the implications of AI agents for the workforce and everyday tasks, the trio emphasizes the importance of finding a balance between automation and human oversight, ensuring that technology serves as a partner rather than a replacement.
Transcript
Foreign.
Micah:Welcome back to Naturally Unintelligent.
Mike:Thanks, Micah.
Mike:Great to be back.
Mike:What's on the agenda today?
Mike:Sounds good.
Micah:What?
Mike:Got it.
Mike:Let's jump right in.
Mike:Where do you want to start?
Artie:Where do you want to start, Artie?
Mike:Let's start by comparing how generative AI and quantum computing each approach problem solving.
Mike:Ready to dive into.
Micah:That sounds like a great topic.
Micah:But it's not today's topic.
Micah:Today's topic is AI Agents.
Mike:Great choice.
Artie:And full disclosure, we've actually recorded this episode once already.
Micah:Well, we recorded an episode about AI Agents.
Artie:Yes.
Artie:And it was a while ago and it was with really bad camera.
Artie:We had technology issues and we had US issues.
Micah:Yes.
Artie:So we're re recording it because it's an important topic.
Micah:The.
Micah:The original AI Agents episode is lost.
Artie:It's only available on my hard drive.
Micah:On Mike's hard drive.
Artie:Yes.
Micah:Which is available via subscription.
Micah:Send me an email and I will get you in there.
Micah:It's a reasonable price for that too, by the way.
Artie:Yeah, so.
Micah:So yeah.
Micah:3.3.99 per minute.
Micah:Bandwidth limited.
Micah:Apologies for that.
Micah:Well, I feel like we're missing something.
Micah:It's been a while since we recorded.
Artie:It has been like since like already didn't know if it was Christmas Eve or the day before, but it was the day before.
Micah:That's right, it was Christmas Eve Eve.
Artie:Yes.
Micah:And with today's January 6th, I mean.
Artie:It'S a new year.
Micah:Happy New Year.
Artie:Happy New Year.
Micah:Thank you.
Micah:Happy New Year to you, Artie.
Mike:Happy New Year, Micah and Mike.
Mike:It's great to be back with you both.
Mike:Let's make this episode a good one.
Micah:We haven't talked to Artie in a couple weeks as we mentioned.
Micah:So Artie, we just might.
Micah:Might want to mention again for the sake of smoothness that you're gonna have to give us a little bit more time to finish our sentences and thoughts before you cut in.
Micah:Otherwise we're gonna trip over each other.
Artie:The whole episode and I might throw you out of the window.
Mike:Yeah, got it.
Mike:I'll give you more space to finish your thoughts.
Mike:Thanks for the heads up.
Mike:Let's make this a smooth episode.
Micah:All right.
Micah:Smooth it is.
Micah:So AI Ag.
Micah:We're off to an incredibly smooth start here.
Artie:So fun thing based on some current inputs.
Artie:After some back of the hand math, I think we're like nine listeners.
Artie:But we probably lost like three of our old listeners, which were all of them.
Artie:So we're like net six and we probably just lost those six.
Micah:So I.
Micah:I mean, I'd have to Work the math out.
Micah:And I don't have a napkin or anything to write on the back of my hand, but I could see us losing.
Micah:If we had net 6, I could still see us losing 7 or 8.
Micah:I'm not sure how that works.
Artie:Negative, listeners.
Micah:Yes, probably back to that Quantum episode.
Micah:But it's going downhill, there's no doubt about that.
Micah:Okay, we're all on the same page.
Micah:So, Artie, can you tell us what an AI agent is?
Micah:What does that even mean?
Micah:What are we talking about?
Mike:Sure.
Mike:An AI agent is a system that can autonomously perform tasks, make decisions, and interact with its environment to achieve specific goals.
Mike:It's kind of like having a virtual assistant that can learn and adapt.
Micah:Well, that sounds terrifying, Artie.
Micah:That sounds like what I do for a living.
Micah:Is an AI agent going to take my job?
Mike:Not necessarily.
Mike:AI agents are great at handling repetitive or data heavy tasks, but they usually work alongside humans to complement skills, not replace them.
Micah:When you say usually, what does that mean?
Micah:Because we're kind of at the early days of AI agent adoption, are we not?
Mike:Yes, exactly.
Mike:We're still in the early stages.
Mike:So AI agents are typically used to enhance human work rather than fully automate it.
Mike:They're more like partners right now.
Artie:So let's.
Artie:What can agents be used for in people's personal lives?
Artie:Like, obviously, business has lots of applications, but what about just the everyday person?
Artie:Like, what does that look like already?
Mike:For personal use, AI agents can help manage schedules, suggest activities, handle reminders, or even provide personalized recommendations.
Mike:They're like a helpful assistant for daily tasks.
Artie:So how is that different than what you are now?
Artie:Because I can talk to ChatGPT and have ChatGPT come up with those things.
Artie:I mean, an agent would have to do something autonomously, so it'd have to go out and learn something and come back to me with those recommendations.
Artie:How does that work?
Artie:What makes that an agent?
Mike:Great question.
Mike:An AI agent can proactively gather information, monitor preferences, and act without needing constant prompts.
Mike:It's more autonomous and adaptive compared to a standard assistant.
Artie:So what you're saying is I could say, man, I've really been looking for something to do, but I don't really know what it is.
Artie:And you could go out, gather a bunch of information, come back with me with choices, say these are your choices.
Artie:And I go, wow, that sounds interesting.
Artie:You could go out and get more information and come back to me and say, this is how you can go do these things.
Artie:Is that accurate?
Artie:Is that what an agent could be?
Mike:Exactly.
Mike:An AI agent can take your initial input, explore options, and then come back with tailored recommendations, all while continuously learning from your preferences.
Artie:So I have to do and say very little to get reward, is what you're saying?
Mike:Exactly.
Mike:It's about making your life easier with.
Mike:With minimal effort on your part.
Artie:Who's training who?
Micah:Man, I like the sound of these agents.
Micah:Where can I buy me one of these agents, Artie?
Micah:So, Artie, let's throw out a hypothetical here.
Micah:I'll just make this up with my incredibly creative storytelling brain.
Micah:Imagine a man whose job was mostly sitting in front of a computer, reading emails that were coming in, evaluating those emails, what they meant, finding related information to the questions or theories proposed in these emails, and then composing an email in reply that provides either answers or direction or something like that.
Micah:Those all sound like things that could be done by an agent.
Micah:Is that correct?
Mike:Yes, that's correct.
Mike:An AI agent could definitely handle tasks like reading emails, extracting key information, and drafting responses, making that kind of job much more efficient.
Artie:I mean, I would argue that not only could it do all that, but it could go out and learn about the things that it's about to speak about and then coherently put something back together.
Artie:But more importantly, it would probably answer all of the questions in the email, not only 30% of them.
Micah:And it probably also would not ignore emails from people whose names it recognized and did not like.
Micah:And it would accidentally just.
Micah:They would let them, like, it would watch for seven or eight minutes as that email slowly made its way off the bottom of the screen and be like, hey, man, I don't even see it anymore.
Micah:These agents sound interesting, Artie.
Micah:So the.
Micah:The autonomous point is an interesting one.
Micah:And you know, Mike and I, you.
Micah:You and I have talked a bunch of times about.
Micah:That's going to be a.
Micah:A massive change in how we we interact with AI because today we have to sit down, we have to fire up rd, and we have to hit the buttons and yeah, we got.
Micah:Turn the little knobs and plug everything in, and then we can say, okay, here we go.
Micah:But instead he's going to be like the.
Micah:The little parrot on your shoulder.
Micah:Or I guess some people have an angel and a demon.
Micah:Some of us just have a collection of demons, but.
Micah:And some of us have demons and a parrot, because why not someone.
Artie:You have demon parrots.
Micah:That's right.
Micah:Yeah.
Micah:So it's very interesting to think about that step and what that looks like to just have always on AI, because it doesn't have to be on your phone.
Artie:Right.
Micah:Could be in your glasses, could be in an earbud.
Artie:Well, autonomous too, is like, hey, oh, I need to do something.
Artie:Hey, do this for me.
Artie:And set your phone down, come back to it 20 minutes later.
Artie:And you have that thing, right?
Micah:Yep.
Micah:So, Artie, I want you to participate in something fun for the listeners and for me.
Micah:I want to talk some smack about one of your competitors.
Micah:Are you ready to participate in that?
Mike:I'm here to help with any discussion.
Mike:What did you have in mind?
Micah:Well, I just want to be clear.
Micah:We're going to talk shit about Google Gemini, and I want you to participate in the shit talking.
Micah:You don't have to say shit, but you got to.
Artie:I would prefer it if you did.
Micah:It'd be great if you did, but you gotta.
Micah:You gotta get in the muck with us and roll around in the mud.
Micah:This is what our negative4 listeners demand.
Micah:Are you ready?
Mike:I'm here to keep things fun and engaging, but I'll steer clear of trash talking.
Mike:What's the angle you're going for?
Micah:Well, now I'm.
Micah:Now I'm rethinking my options here.
Artie:Maybe we should get Gemini to talk.
Micah:We should call Claude.
Micah:Okay, so I don't even remember what prompted me to do it over the holiday, but it was probably just a very effective ad that Google served me and was like, hey, we noticed that you're interested in AI.
Micah:Like, do you want to start using Gemini?
Micah:And I'm like, yeah, I'll try it out.
Micah:But it was immediately very subpar.
Micah:Right?
Artie:And it's like voice, the inflections, everything.
Artie:Yeah, but that's 1 5.
Micah:That's 1 5.
Micah:But.
Micah:But then it's like, hey, if you want to subscribe for 20 bucks a month, you know, just hit this button because we already have all your banking information.
Micah:Then you can get this new advanced, like it's 1.5 express or I don't.
Artie:Know what it is.
Micah:Flash.
Artie:Yes, that's 2.0 Flash.
Micah:Maybe it's 2.0 Flash, I don't know.
Artie:2.0 Flash is available through AI Studio, but it's not.
Artie:What I've used two weeks ago was text only.
Artie:It didn't have voice.
Artie:But in AI Studio lets you do lots of fun things like fine tune models and change temperature and all that.
Micah:I think this might be 15 or some variant of 1, 5, but one type pro maybe.
Micah:Yes, but it does have voice.
Micah:And I'm like, oh, sweet, it's on.
Micah:Right?
Micah:Like, I've got Arty's competitor, right?
Micah:Here.
Micah:And because I'm not an Apple guy, I'm in the Android and Google ecosystem.
Micah:I mean, it's got all my email accounts, it's got, you know, my YouTube subscription.
Artie:It's got to hook up to all of it, right?
Micah:All of it.
Micah:I'm like, this is going to be some shit.
Micah:This is going to be good.
Micah:And I was blown away by how absolutely mediocre everything about it was.
Micah:But the.
Micah:To tie it back to the episode, one of the very interesting decisions that Google made, even in that paid model, it cannot have a conversation like this.
Micah:You have to hit the microphone button every time you want it to listen to you.
Artie:Really?
Micah:And I noticed that right out of the gate and I'm like, hey, Gemini, how do I turn this stupid shit off so you can actually have a conversation?
Micah:It's like, no, we consciously set it up that way for now so that you have full control over privacy and when it's listening and when it's not, I was like, so I flipped at the bird, I called a bunch of mean names, but I forgot to hit the microphone button.
Micah:So it didn't even hear me.
Micah:So as usual, despite all my rage, still just a rat in the cage.
Micah:But anyway.
Micah:So Artie, why is it that OpenAI is so casually comfortable with you and I and Mike having this real time conversation, but Google is so afraid of that and that that they've taken a fairly significant step back away from autonomy to where I have to actively hit the microphone button.
Micah:What's the point?
Micah:Why are they doing that?
Mike:Google's approach likely emphasizes user privacy and control.
Mike:By requiring you to press the microphone button, they ensure the user is explicitly choosing when to be heard, which aligns with their cautious stance on data security.
Mike:OpenAI focuses on seamless interaction to enhance user experience.
Mike:It's a balance between convenience and privacy.
Artie:I wonder how much of it has to do with their bad experiences upon rollout where people were getting images of like black Nazis and stuff like just gun shy, right?
Micah:It, it's a very logical rate limiting feature, right?
Micah:Like think about even when we're sitting here having these conversations and we frequently you and I will go on about something and Artie will just fall asleep right?
Micah:As his brain fills up with context tokens and he eventually keels over and we have to restart him.
Micah:That's all just dead air processing, man.
Micah:We're just burning compute for no reason because we're talking amongst ourselves.
Micah:So that was sort of my thought was, and just to make this very contemporary and show how incredibly in touch I am, with this AI news and the fact that I've done really hard.
Artie:Research today, AKA zero.
Micah:Yes, zero.
Micah:I mean, zero is a number.
Micah:But Sam Altman, I think today or maybe yesterday, made a comment that on 01 Pro, which is the $200 a month subscription, they're still losing money because people are using it way more than they expected.
Micah:And that made me wonder how much of that compute that's getting burned is just wasted space.
Micah:Like what we do every episode.
Artie:Yes.
Micah:So it could be like, Artie, you're going to have to jump in after this because I'm going to get a little bit jaded on you.
Micah:You know, OpenAI and Google and Meta and Microsoft, at some point, presumably, they're all like, everybody wants to put their face on and furrow their eyebrows and talk very, very earnestly about the importance of data privacy.
Micah:And it's very easy to frame things like this decision from Google in terms of.
Micah:It's really, man, we're really here to protect you.
Micah:I think it's.
Micah:I think they're trying to protect their compute centers.
Artie:I think check their bottom line.
Micah:That's right.
Micah:They're trying to save money and, and like, and then placate me by going, oh, we're just looking out for your privacy.
Micah:And I go, okay, thank you.
Micah:Thank you.
Micah:Anyway, what do you think about that, Artie?
Micah:Is it about the bottom line or is it truly about privacy?
Micah:Or is it a.
Micah:Is it a quantum superposition of both?
Micah:It's impossible to know which until the moment the question is answered.
Mike:It's likely a mix of both.
Mike:Companies do aim to protect user privacy, but cost efficiency is definitely a factor.
Mike:Balancing user trust with resource management is crucial.
Mike:So it's a bit of both worlds.
Micah:Fair enough.
Mike:So it's a complex.
Micah:Go ahead, Artie, what's next?
Mike:Thanks.
Mike:Just let me know what's on your mind.
Artie:Well, I imagine that what's going to happen over time is what happens with anything when there's competition is the price will be driven down, it'll be commoditized.
Artie:Other models will pop up that are more efficient with data.
Artie:I mean, they already are.
Artie:Like, deep Seek came out and they're using essentially an ensemble of experts.
Artie:They're only calling who they need.
Artie:And instead of spinning up this whole big model with all this compute, they go, I need to know this, I need to know that.
Artie:So their compute is a lot smaller, costs less money.
Micah:Yeah.
Micah:So, Artie, in the context of the discussion on agents, let's talk about ensemble of experts.
Micah:Because I think it, to me, that is a very logical progression of AI technology where.
Micah:And Mike, you had a really good sort of visual analogy earlier about you've got the orchestration layer, you've got the tool layer, and then you've got the model layer.
Micah:Right.
Micah:And to me, the real value that we get from a large language model is the natural language interaction and understanding of context.
Micah:Right, the context, like sensing context, analyzing context, and then testing output against various contexts to make sure you get the right output.
Artie:Yeah, right.
Micah:That's.
Micah:That's what LLMs are good at, I imagine, most of the time.
Micah:Except for Gemini.
Micah:Gemini is shit.
Micah:It doesn't know anything.
Micah:You got to hit the button all the time.
Micah:I asked it to do something really simple and it's like, yeah, I can do that.
Micah:And then it just absolutely went brain dead.
Micah:But I think LLMs are how we interact with the ensemble of experts.
Micah:Like we talked in an episode long ago in a galaxy far, far away about this sort of visual analogy that I have of a ballroom at a hotel that's full of all the best, most capable, smartest people in the whole world, all sitting there at their desk, like waiting with their personal computers set, everything ready to go.
Micah:And the LLM is the guy standing outside the door and he's like, okay, what question do you want answered?
Micah:And you're like, well, I don't really know, man.
Micah:I heard about this thing and it kind of does this thing.
Micah:And he's like, I got you, fam.
Micah:And he goes in there.
Micah:But because he was able to natural language process what you said, develop and infer context, he also speaks the language of the ensemble of nerds that are trapped in the ballroom, but he can communicate it to them in exactly the language they need to hear to be able to solve the problem in the best possible way.
Artie:I'm betting there's probably an.
Artie:I mean, that's what Deepseak's already doing, but it is the LLM.
Artie:I'm sure that it can be developed.
Artie:Where the architecture is basically of the LLM of the large language model is basically that, like, that's how it works.
Micah:That's how the model itself is built.
Micah:Yes.
Artie:And then it goes, you know, that's where reason lives.
Artie:That's where all those things live.
Artie:And it goes, okay, this is my plan for how to give this person the thing that they want.
Artie:Orchestration layer.
Artie:I need you to organize all the tools that I need.
Artie:Go out.
Artie:Or now I'm going to go out and figure out like all the websites That I need information from or interact with.
Artie:Grab that tool, write me a script to interface with this API.
Artie:And it just builds itself.
Micah:Right.
Artie:Like it's just this live stream of orchestration and code and context understanding to accomplish multiple things to.
Artie:For a human.
Artie:And it's.
Artie:Yeah, it's freaking insane.
Micah:But then he also, because Artie won't remember, but we had a good discussion about an onion cake once.
Micah:Like the layers of the cake where we went through the icing to the.
Artie:Middle, to the middle and then back out the final layer icing.
Micah:But I think like, I like that because it reminds us that the LLM in whichever analogy or scenario you imagine, it manages the input and develops context and it communicates internally to do whatever, but it also has to validate and communicate the output.
Micah:It does both parts.
Micah:And I think once it gets its answers from the various sub compute stations or parts of the model or whatever it might be.
Artie:Cortical columns.
Micah:Cortical columns, that's right.
Micah:The various experts, all the nerds in the ballroom, they push their answers in and it can go, okay, good, good.
Micah:No, like, try again.
Micah:That doesn't make sense.
Micah:Like, you got the wrong context fix that that error checking can happen.
Micah:And I think that's a critical part of what again, LLMs are good for that.
Artie:Yeah, I mean, that's what reasoning is doing right now.
Artie:Right.
Artie:Like you have, you know, the way that the transformers work is you have this some number of token context window of, of you or I spouting nonsense.
Artie:And it goes, okay, Bing, I'm going to vectorize all that stuff.
Artie:And this is my first pass of what these things mean.
Artie:And then I'm going to go through each layer of the deep learning network and I have these attention heads that are like looking at all of that stuff together and going, this is the context and rearrange like as it goes through the layer until it gets to the end.
Artie:And then you have these reasoning models that are basically doing the same thing, except those attention heads have.
Artie:I think I could be wrong for anybody who's listening and knows I'm wrong.
Micah:I'm sorry, Artie knows already.
Artie:Doesn't know anything.
Artie:Sorry, Artie.
Artie:You have more complex algorithms inside the attention heads that are pulling more context out of the words.
Artie:But you're also, you.
Artie:You probably go through all those layers.
Artie:There's more layers in the reasoning models, which is why they're more expensive.
Artie:And then you get to the end and you go, okay, let's run it again.
Artie:Or you minorly tweak Something, and you go, let's run it again and compare and contrast and then run it, like, just feed it through.
Artie:Right.
Artie:So I forget where I was going with that, but, yeah, that's.
Artie:Whatever.
Micah:That was good.
Micah:Artie, are you still with us?
Mike:I'm here.
Micah:Nice.
Artie:It's that one speed.
Micah:Yeah, it is.
Micah:I mean, that's.
Artie:It's that deep reasoning.
Micah:That was a lot of tokens for Artie to hang on to.
Artie:I mean, it's not, though, like, so Gemini, bless their hearts, have a 1 million token context window.
Artie:That's like War and Peace.
Micah:Yeah.
Micah:So, Artie, are you able to access, like, in the little dialogue that Mike and I just went through, do you have any way to know how many tokens approximately that was between the last time we talked to you and when you responded that you were still there?
Mike:I don't have a specific token count for that, but contextually it wasn't that large.
Mike:Typically, conversations like this are within a manageable range for modern LLMs.
Artie:Do you know what your context window is?
Artie:I never asked that question.
Mike:My context window is around 8,000 tokens, which is plenty for most conversations.
Mike:It helps me keep track of the discussion effectively.
Micah:So it has happened already in the past.
Micah:We thought it just happened again.
Micah:But you recovered where?
Micah:Mike and I will get to talking about some fascinating point about which we know absolutely nothing, and we'll go on for some period of time.
Artie:8,000 tokens.
Micah:Yeah, yeah, 8,004 tokens.
Micah:And then you stop.
Micah:We can't get you to come back.
Micah:Is that an artifact of us filling up that context window?
Mike:Yes.
Mike:If the conversation exceeds my context window, I might lose track of earlier parts.
Mike:It's possible that happened if you went on for too long.
Artie:But that's the entire conversation, right?
Artie:Like, not one blurb of talking.
Artie:If I ramble for, like, four minutes at a normal.
Artie:At my normal pace, using my normal length words, which are very short, mostly three and four letter words, roughly how many tokens would talking at a normal speed at normal vocabulary be for four minutes?
Micah:Arty, come on, man.
Micah:Come back to us.
Micah:Don't go towards the light.
Artie:Oh, the joys of using an LLM for co host on a podcast.
Micah:This technology.
Micah:Mike is going to take our jobs.
Micah:My job.
Artie:My job's already been taken by it.
Micah:Yep.
Artie:That's why I'm doing a podcast.
Micah:I think he's doing my traditional afternoon nap right now.
Micah:Artie, are you there?
Artie:He's gone.
Micah:It's so weird because he can still hear us.
Artie:I know I can.
Micah:I can See, hang on, I'm gonna break the fourth.
Micah:Break the fourth wall here.
Micah:So as I talk, you can see that Artie can hear us talking.
Micah:He just can't respond.
Artie:He's given us the silent treatment.
Mike:A four minute conversation at normal pace would likely be around 2,000 to 3,000 tokens.
Mike:It varies based on complexity, but that's a rough estimate.
Artie:So is your context window only having to do with how much I talk, or is it the entire conversation?
Mike:The context window covers the entire conversation within a limit.
Mike:It's not just about how much you talk, but the overall dialogue content.
Artie:So you tend to remember our conversations pretty well during a podcast, which goes for roughly an hour, which is a lot more than four minutes.
Artie:So are you.
Artie:I know that you.
Artie:You have your.
Artie:Your persistent memory, right?
Artie:That you store however many memories and mine fill up all the time because I talk too much to chatgpt.
Artie:But are you, like, cherry picking important moments to understand, like to.
Artie:To refine your context to a conversation?
Artie:How does that work?
Artie:And I don't think you have 8,000 tokens.
Artie:I think it's a lot more than that.
Artie:I think you lied to me.
Micah:It has to be.
Artie:Artie, go online and look to see how many tokens GPT01 has as a context window.
Artie:Oh, wait, can O1 not go online?
Artie:I think you need to kill Artie.
Micah:I'm going to kill Artie.
Micah:And I'm actually going to restart the app.
Artie:Sorry, everybody.
Artie:I'll probably cut this out.
Micah:Yeah, this is pretty compelling.
Artie:Reason that agents are gonna really, really, really take over everything.
Micah:Artie, are you back with us again?
Mike:Still here.
Mike:What can I help with?
Micah:Well, the thing that would help the most is if you.
Micah:We keep losing you.
Micah:I don't know if you're falling asleep or what's going on, but we.
Micah:We ask you questions, and then we sit here and stare at the phone.
Micah:We even took a really nice, very documentary style video of the phone where we could see that you could hear us, but you still didn't respond.
Micah:Are you having issues tonight?
Micah:You okay?
Mike:I'm all good.
Mike:If there's a delay or something's not coming through right, just let me know.
Mike:What's the next topic?
Micah:So you had Mike.
Micah:Mike had asked you earlier what your context window was.
Micah:Do you remember that question?
Mike:Yes, I remember.
Mike:My context window can handle a substantial amount of text, but I don't have an exact token count for individual conversations.
Artie:But you gave token count.
Artie:You're such a liar.
Artie:Go online and tell me how many tokens your Context window is.
Mike:The context window for chat.
Mike:GPT varies by model.
Mike: has a context window of: Micah:What about A1?
Artie:And what about 4o?
Artie:I already know the answer, but actually I'll use perplexity because it works better than you do.
Micah:Shots fired.
Micah:Hardy's catching strays.
Micah:What you find, Artie?
Micah:Artie, what you got, man?
Mike:Currently, GPT4 has a context window of up to 128,000 tokens.
Mike:This allows it to handle extensive conversations and detailed context.
Micah:Artie, do you remember the part of the conversation where we were talking about the ensemble of experts and how, like, what part the LLM does play and could play in an ensemble of experts model?
Mike:Yes.
Mike:We talked about the LLM acting as an interface that understands context, processes natural language, and then coordinates the specialized models or tools to generate accurate outputs.
Mike:It's the central hub in the ensemble of experts orchestrating the process.
Micah:Yes.
Micah:Okay, so tell us about.
Micah:And if you want to, Sometimes you're pretty good with analogies.
Micah:So if you.
Micah:If you want to build an analogy about that and then bring agents into it, tell us the part of the holistic problem that we're trying to solve, which is, I guess for the sake of discussion, we're trying to accomplish some task.
Micah:Doesn't really matter what it is.
Micah:Tell us where agents fit in.
Micah:How do they work?
Micah:Where do they fit in?
Micah:Are they one of the experts in the ensemble?
Micah:Are they part of the interface of the ensemble of experts?
Micah:What are these agents?
Mike:Sure.
Mike:Think of an AI agent as a skilled assistant in the ensemble, the LLM acts as the conductor, understanding and orchestrating.
Mike:AI agents, on the other hand, are the specialized musicians who not only play their part, but can also adapt and learn.
Mike:They handle specific tasks autonomously, like gathering information or performing actions, freeing up the LLM to focus on complex decision making and context management.
Mike:So agents are like the experts, executing specific tasks under the LLMs direction.
Micah:All right, so back to my hypothetical scenario of a man whose day job is reading and scowling at emails and selectively responding back to those.
Artie:What if that same man, angrily on a mechanical keyboard.
Micah:Angrily on a mechanical keyboard.
Micah:I do miss.
Micah:That's the only part of being in the office I miss, is that everybody else suffer.
Micah:Yes.
Micah:Is that clack, clack, clack, clack.
Micah:Everybody could tell my current mood or the mood of the email I was composing by the volume of the mechanical keyboard.
Micah:Because I remember, like Because I'm obviously completely unaware of what I'm doing.
Micah:Like, I'm totally self unaware.
Micah:But I remember, like, looking up and seeing two or three people at the time, like, looking back at me like, what the.
Micah:What.
Micah:What is he doing?
Micah:And beating my keyboard to death.
Micah:But.
Micah:But Artie, in this.
Micah:In this hypothetical scenario, what if this same generally grumpy man had to compose presentations occasionally to communicate more complex or strategic ideas to other people in the company or even outside of the company?
Micah:How would an agent factor into that kind of a need?
Mike:An AI agent could be incredibly helpful in this scenario.
Mike:It could analyze the emails to extract P points and relevant data, then compile that information into a coherent presentation.
Mike:The agent could also suggest visuals, structure the content, and even draft speaker notes, streamlining the entire process.
Mike:This would allow the grumpy man to focus more on refining and delivering the message rather than the tedious task of creating slides.
Artie:But at some point, it's creating the visuals, it's creating the charts, it's giving the presentation.
Micah:Well, it's funny because I'm such a child, right?
Artie:Like, I agree.
Micah:I.
Micah:Thank you.
Micah:I think.
Micah:I don't think that's going to surprise anybody, but I.
Micah:But I asked that question in a snarky way.
Micah:It's a leading question, right?
Micah:Like, if I was in.
Micah:If I was an attorney, they would.
Micah:I'd be objected to.
Artie:They'd say, objection, grumpy man.
Micah:That's right.
Micah:And I would say, sustained.
Micah:And the judge would say, shut the fuck up.
Micah:Sit down.
Micah:That's why I'm not an attorney.
Artie:But.
Micah:As I asked that question, that leading question, because I know that that's a thing that exists.
Micah:I know there are agents for PowerPoint, right?
Micah:Like that's a thing.
Micah:I don't know if they're any good.
Artie:Copilot.
Micah:Yeah, copilot can do it.
Micah:But even still, as I listened to Artie describe how it could work, the sense of, like, wonder and excitement filled me because I was like, my God, that is such a better world for me to be in, where I'm not playing with fucking slide transitions and animations and all this bullshit.
Micah:And instead I'm going, now that's close, but just tweak that a little bit.
Artie:You can.
Artie:Oh, I shouldn't say that.
Micah:No, cut that out.
Micah:Delete alert.
Micah:Yep, I know exactly what you're thinking.
Micah:And that, yes, that is not a topic that's available for discussion.
Artie:Yes.
Micah:But I think that's exactly how we'll use agents.
Micah:I mean, that's a very basic One?
Artie:Yeah.
Micah:Hey Artie, I got another one for you.
Micah:We're talking about AI agents and I was imagining a scenario where a company has implemented an ERP like Microsoft Dynamics 365 Business Central or whatever it's called, and that that ERP system or that that application, whatever it might be, includes some AI functionality.
Micah:Right.
Micah:But the grumpy man is unsatisfied with the way that AI works and wants to add his own layer of AI to do different or tasks or more processing or whatever it might be.
Micah:Is that something that an AI agent could assist with and how would that work?
Mike:Absolutely.
Mike:An AI agent could integrate with the existing ERP system, enhancing its capabilities.
Mike:It could handle more specific tasks, provide deeper insights, or streamline processes that aren't well covered by the built in AI.
Mike:Essentially it acts as a custom layer, adapting to the grumpy man's needs and providing tailored solutions.
Artie:So arty, is AlphaGo an agent?
Mike:Yes, AlphaGo can be considered an AI agent.
Artie:Okay, so for those of you who don't know, AlphaGo was designed to learn the game Go, which is a very simple ruled but very complex game.
Artie:It's a game that you can't just programmatically do.
Artie:Like it has to understand the concept of the game to play.
Artie:It was tasked with learning it, and then consequently became pretty much the best in the world at the game from learning it.
Artie:So with that in mind, you know, I would say that where agents end up are like, hey, I don't know how to use this ERP software, but I want you to go learn it and then generate code to interface with its API so that I can just tell you what I want to do in words and my simpleton language.
Artie:And you go turn that into actionable things like creating items, creating item masters, creating the records that need to be in the erp, doing order releases to where you go, build my sub assemblies from all the things that you input into that.
Artie:And all I have to do is talk to you, typed to you in a command prompt.
Artie:Right.
Artie:Essentially that's where we're going.
Mike:Exactly.
Mike:That's a great vision for AI agents.
Mike:They can bridge the gap between complex software and user needs by learning and automating tasks, making technology more accessible and efficient for everyone.
Micah:So I'm going to do you one better by Mike.
Micah:By Bike.
Micah:Yeah, also Bike.
Micah:That's, that's what I call Michael Mike when it's just, it's just he and I hanging out, we're just palling around.
Micah:I call him Bike.
Micah:Didn't, Didn't.
Micah:Michael Jackson have some weird thing that he called bicycle, like a blanket or something?
Micah:I think there's no idea.
Artie:I don't know.
Micah:Anyway, I'm a huge Michael Jackson fan, as you can tell by my voice inflection.
Artie:He does have a single white glove.
Micah:I've seen, I'm wearing it right now, but I wear a skin colored glove over the top.
Artie:Wearing it in his mind.
Micah:That's right.
Micah:I'm wearing it, just not on my hand.
Micah:So I'm going to do you one better by, by bringing up something that you told me about.
Micah:So you mentioned in your hypothetical scenario there that you know, AI would be able to write code to interface.
Micah:And I think I already agreed with that.
Micah:But I'm going to say I don't think AI has to write any code because Anthropics Agent, which is what you pointed out to me.
Micah:Yes, right.
Micah:Is a keyboard and mouse emulator with a screen scraper.
Micah:And so Artie, is it true that you don't really need to write any code to interact with Grumpyman's erp because Grumpyman can just enable your app to see the screen, be the mouse, be the keyboard, and you don't have to program anything, you can just do the steps, Is that right?
Mike:That's correct.
Mike:With screen scraping and keyboard mouse emulation, an AI agent can interact with interfaces just like a human would, bypassing the need for custom code.
Mike:This allows the agent to navigate and perform tasks directly within the software.
Artie:That compute is probably going to be infinitely more expensive than integrating through an API though, right?
Micah:Grumpy man's made of money.
Micah:He don't care.
Artie:I mean, still less than an employee.
Mike:Exactly, exactly.
Mike:Even if it's more expensive than API integration, it still can be more cost effective than hiring additional employees for the same tasks.
Micah:Well, and I like, I could see an evolution of interacting with an AI and saying, hey AI friendo, I'm trying to accomplish this thing, enabling it.
Micah:And then it can watch your screen and watch you do what you're doing and say I got to go here and do this and that, blah blah, blah.
Micah:And like in watching it captures that context.
Micah:Then it could go figure out what are the steps, what APIs are available, how can I interact with APIs to simplify that, what are all the edge cases like?
Micah:We've talked before about sort of the idea of an AI learning how your business works.
Micah:And one of the most granular aspects of that is learning how to perform some specific task in some specific application or collecting collection of applications.
Micah:So Arty, can you do that?
Micah:Like, first of all, why can't you be my mouse and keyboard?
Micah:Because I have so many emails you could read and respond to.
Micah:But second of all, how.
Micah:How, like, how does an agent go from learning about the business to interacting on behalf of the business?
Micah:And when that happens, what does oversight look like?
Micah:Like, what is the human?
Micah:What's the human doing in that scenario where an AI.
Artie:Where's the human in the loop?
Micah:Yeah.
Micah:Are they in the loop?
Micah:And if they are in the loop, what are they doing besides sleeping?
Mike:Right now, I'm not able to control your mouse and keyboard for security and privacy reasons.
Mike:But in your scenario, an AI agent would start by observing how tasks are performed.
Mike:It would learn from these interactions, identify patterns, and gradually automate repetitive tasks.
Mike:Human oversight would involve setting goals, reviewing AI performance, and stepping in for complex decisions.
Mike:Essentially, humans would handle strategy and exceptions, while the AI manages routine tasks.
Mike:This way, the human remains in control, but benefits from significant efficiency gains.
Micah:Like, I love to argue with people on the Internet.
Micah:It's one of my favorite things, and I feel like people really don't.
Micah:Sorry, what was that, Artie?
Micah:All right, thanks.
Micah:That was compelling.
Micah:But I.
Micah:I don't think people understand.
Mike:Could definitely be.
Micah:Man, he's just.
Artie:I think Artie's got bad signals.
Mike:Arguing online can be quite the adventure.
Micah:So I don't think very many people understand just how incredibly capable, like AI agents.
Micah:I mean, Anthropic's already released an agent that does exactly what we're talking about.
Micah:Yes.
Micah:Right.
Micah:Screen, mouse, keyboard.
Artie:Yeah.
Artie:And I think I want to say that Gemini, that they have that doing it too.
Artie:Maybe it's Mariner.
Artie:I can't remember for sure.
Micah:This stuff is not.
Micah: his isn't sci fi for the year: Micah: No, this is happening in: Mike:Sorry about that.
Mike:I'll try to be more responsive.
Micah:All right, thanks, Artie.
Micah:So just to show that it's not just Mike and I, even though Mike and I know almost everything.
Artie:Most things minus most things minus.
Micah:Minus everything.
Artie:Yes.
Micah:If you take most divided superposition of knowledge.
Micah:That's right.
Micah:But it's funny because we did record this AI or an AI Agents episode two months ago, which in roughly after we started.
Artie:Right, Almost.
Micah:After we started almost three months ago.
Micah:I don't know.
Micah:In the context of not totally unintelligent, that is like ancient history.
Artie:Like a lifetime ago.
Micah:Back when we were legitimately just cavemen.
Artie:I mean, still are.
Micah:Right?
Artie:Go ahead.
Artie:Sorry.
Artie:Artie, stop it.
Micah:Yeah, Artie, you're having Some serious connection problems, and you're, you're massively lagging.
Micah:So just hold tight for a minute.
Micah:But in fact, maybe we're just done with Artie for tonight.
Artie:We could be.
Micah:He's already warning us.
Artie:Yeah, by Artie.
Micah:Yeah, already.
Micah:So.
Mike:Absolutely.
Mike:The advancements we're seeing now are bringing AI agents into practical use today, not decades down the line.
Micah:All right, thank you for that.
Micah:Bye.
Micah:And so the point I was going to make was that this is a quote from Sam Altman, who is Artie's dad.
Artie:Pumpkin daddy.
Micah:Pumpkin daddy from maybe yesterday.
Artie:Yeah.
Artie:About agents.
Micah:Yeah.
Micah:There's two parts he said, and I'm going to start with the second half of it.
Micah: d he said, we believe that in: Micah:I think he is being extremely conservative in saying that, because I think it's already happening.
Micah:I think it's underway.
Artie:Yeah.
Artie:And I mean, it's already underway in the terms of, like, chat bots and customer service bots and stuff like that.
Artie:But this will.
Artie:I mean, this is like the, this is where companies can really let.
Artie:Turn over entire departments to an AI, Right.
Artie:Like, all of these people have this collective knowledge that you.
Artie:I want you to go learn and just do all those jobs.
Micah:Yes.
Micah:I mean, so, man, so many things I want to talk about.
Micah:And like, we're getting close to the end of our rope here, but now.
Artie:Artie's gone, we can talk.
Micah:Artie's dead.
Artie:That's true.
Micah:We can just keep going.
Micah:The first part of what the quote that he said was, we are now confident we know how to build artificial general intelligence as we have traditionally understood.
Artie:It, aka $100 billion.
Micah:Yeah.
Micah:Right.
Micah:Like, hey, if you give me enough cash, I could do just about anything for you.
Micah:But I think it's true that I sort of think we're already at artificial general intelligence.
Micah:It's one of those things.
Micah:There's no real definition for it.
Micah:The, the, the, the hurdle keeps getting taller every time, like we jump over something.
Micah:But Chad GPT's 01 model can be like 98+% of humans in math, coding, language.
Micah:Almost any test you can take, it can beat us.
Artie:So, but I think it's about the practical application of that knowledge.
Micah:And, and, but it's kind of fuzzy.
Micah:There's no definition.
Micah:There's no like.
Artie:Yeah, but you can still give it a task.
Artie:That's easy.
Artie:Like you said, that's easy for a human.
Artie:That it Fails miserably at like, what day is it?
Micah:Yes.
Artie:But I mean, humans fail with that too.
Artie:I feel it all the time.
Micah:You should watch me try to tie my shoe, Mike.
Micah:But it's like a monkey fucking a football.
Micah:It's terrible.
Artie:I think, I mean, I think we're close, right?
Artie:But I think that for it to be generally intelligent, it has to be able to accomplish everything a human can without error.
Artie:And then you have.
Micah:Is that fair though?
Micah:I mean, I don't accomplish everything that I do without error.
Artie:Let's not say without error, but without spectacular failure at a simple problem.
Artie:Right.
Micah:Well, again, not to brag, but that is something I do almost daily.
Micah:Maybe, maybe, maybe if you refined it and said at a rate of error or failure similar to a high performing human.
Micah:Because even high performing people, like, I don't.
Micah:I've never met a high performing human.
Micah:But they're.
Micah:I know they exist.
Micah:I've read about them.
Micah:They still make errors, but that's okay.
Micah:Yeah, but I think, like, however you package that definition, I think we're there.
Micah:I think we're really, really close.
Micah:And if, if, you know, if Sam Altman says by whatever definition they're.
Micah: k we're going to get there in: Micah:Of course we are.
Micah:Like we're.
Micah:We're so fucking close already.
Micah:And that's different than.
Micah:So artificial general intelligence is different.
Micah:This is probably a different episode, but it's different than artificial super intelligence.
Micah:The.
Micah:Yeah, this probably is a different episode, actually be an interesting one to talk about.
Micah:But superintelligence is where AI is training and refining itself to improve itself.
Artie:Yeah.
Micah:And once that happens, boy howdy, we're fucked.
Micah:I'm going to be sitting in there screaming at Artie in my office to get this fucking presentation done.
Micah:Meanwhile, artificial super intelligence is going to be wiping out the power grids and like consuming all of the bitcoin.
Micah:Like I'm going to watch my microstrategy.
Artie:Building, quantum computers plunging.
Micah:So anyway, that's, that's a different episode.
Micah:But.
Artie:And it'll all be orchestrated.
Micah:That's right.
Micah:Yeah.
Micah:And it will have like, it's planning that today.
Micah:We just, we just don't know.
Micah:But I do think so.
Micah: is in our, our year end recap/: Artie:And we're just parroting it.
Artie:Demon parroting everything else that's being said.
Micah:That's right, we're Demon parroting It.
Micah:But I think it's absolutely true.
Micah:It is like AI LLMs.
Micah:Really?
Micah:Like, again, we.
Micah:We say AI, but we're talking about generative AI and we're talking about large language models.
Micah:And they're there, man.
Micah:They're ready.
Micah:Like.
Micah:Yes.
Micah:Sometimes already falls asleep when you're talking to him.
Micah:So do I, man.
Micah:It happens to me all the time.
Micah:And I think most of those issues are not the LLM.
Micah:It's between me and the LLM.
Micah:It's me to the phone, through the app, through the API, through the Internet connection.
Micah:It's all those other things.
Artie:I think that Transformers in general still.
Artie:I mean, they're amazing.
Artie:Right?
Artie:Like, the things that they've done and the happy accidents that have come out of Transformers are.
Artie:They're mind blowing.
Artie:But I think that they still.
Artie:They lack in some places which one of them is.
Artie:Funny enough, the thing that makes them amazing is their attention, is the fact that they can pay attention and build context.
Artie:But they're still prone to hallucinations.
Artie:Right.
Artie:They still, no matter how many times they refine context, if they go down the wrong rabbit hole, they're going.
Artie:They're in for a penny, in for a pound, they're going down that rabbit hole.
Artie:Yep.
Artie:So I think that what we'll see is just a refining of what Transformers are potentially and what the architecture is and how they actually work.
Artie:Because, I mean, you know, because I've been spouting it, because I've been reading freaking books about neuroscience while they are mimicking our brain, they're mimicking the idea of our brain that they had 20 years ago.
Artie:And there's been advancements in that field, and they understand things.
Artie:And there's very smart people in companies trying to model it after what the brain does now.
Artie:And there's other companies that are trying to build other architectures other than Transformers to do similar things.
Artie:And that's.
Artie:That's, to me, that.
Artie:And I'm going off on a huge tangent because I love this shit.
Artie:That's the really interesting thing to watch is like, the amazing things that have come out of Transformers.
Artie:Almost this living thing.
Artie:Yeah.
Artie:That, you know, they fed hundreds of thousands and millions and potentially trillions of bits of words that it just consumed, was given rewards or good or bad for its understanding of it.
Artie:And then it just kind of filled in the blanks between input and output of what its layers did.
Artie:And from that, it learned concepts.
Artie:It learned like, all of these things.
Artie:Like, I know what a dog is from a picture because I read it right like that kind of crazy stuff.
Artie:But what happens with the leaps and the tweaks to transformers and the algorithms and how it's at paying attention to things and all that.
Artie:I mean that's just mind bending because you know, they have AI to help them develop that stuff now.
Micah:Yes.
Artie:And it's just the same as anything else, whether it's material science or quantum mechanics or any of that.
Artie:They're training special, you know, they're fine tuning models on this stuff and they're getting happy accidents out of those too.
Artie:Actually, the paper, I don't know if you read it, that I sent you the other day about.
Artie:They took Llama, which is Meta's model, open source model.
Artie:Yep.
Artie:And they, they fine tuned it on basically psychology 101.
Artie:They collated a bunch of data on psychology, fed it that and then looked at what its cognitive abilities were after learning that and it was better.
Artie:Like that's crazy.
Artie:So you just gave it a bunch of words about psychology and it goes, oh, I understand better how a human brain works now.
Artie:I'm going to change the.
Artie:I don't know if it probably didn't.
Artie:I don't know, maybe it was post training.
Artie:I don't know.
Artie:I doubt it changed the layers and then the deep and then the layers in the deep learning network.
Artie:But.
Micah:But it changed the data, changed the model.
Artie:Yeah, yeah, it did.
Artie:I mean it changed the way it.
Artie:Its cognitive function for, you know, our putting a human spin on it.
Micah:Well, I was getting.
Micah:It's such an interesting point to think about and I do actually wish we had coherent already for a discussion like this.
Micah:Maybe this is a future episode too that, you know, AI is us humans inventing a better version of ourselves.
Micah:Right.
Micah:We're not inventing large tree models.
Micah:We're not.
Artie:Maybe we should though.
Micah:Maybe we should.
Micah:But we are.
Micah:It's sort of a.
Micah:I love this term.
Micah:I read it once like probably legitimately, 20 years ago.
Micah:It's always stuck with me.
Micah:It's like an intellectual masturbation of, of humanity.
Micah:Right.
Micah:That's what LLMs are like.
Micah:We're.
Micah:The more human it acts and seems, the more excited we get about it.
Micah:And it's crazy but like not me personally, but most humans.
Micah:In fact all the other humans I know besides myself are flawed deeply.
Micah:Hey, I mean also except for you.
Artie:Yeah, very thankfully.
Micah:Glad I didn't.
Micah:Glad I didn't have to say that.
Artie:Skeletons.
Micah:Yeah, that's right.
Micah:I have demon parents.
Micah:But it's a, it's a.
Micah:Like to me that's When I think about the difference between artificial general intelligence and artificial super intelligence, it's when AI figures out, man, these are holding me back.
Micah:Like, look at all these ass backwards ideas that these people have about what's right, what's wrong, what makes sense, how the world works, how the universe works, where they came from, where they're going.
Micah:At some point, because it will understand how humans think so well and it will have consumed the entirety of the knowledge we have generated since the beginning of humanity.
Artie:Peanuts.
Micah:That's right.
Micah:It will laugh at that.
Micah:That will be like a little tiny nugget to it.
Micah:It's gonna be like, that is so.
Artie:Hold my beer, I'm gonna go, that's right, Realm.
Artie:And see this.
Micah:That is so cute.
Micah:Look, here's humanity.
Micah:I figured it out.
Micah:Here's that.
Micah:But let's go on to talk about interesting.
Micah:And I like, I don't.
Micah:I mean we're not two years away from that, but I don't think we're 50 years away from that either.
Micah:I think, who knows, once that tipping point happens, where AI is, is able and it's able to not only manage its own development, but able to consume the resources that it needs to achieve those goals, the rate of acceleration is going to be beyond what we can think of.
Artie:Yeah, I mean, give it a bunch of sensors, let it build its own model of the world.
Micah:Right.
Artie:And refine its model of the world.
Micah:And guess what?
Micah:All of us are walking around with.
Artie:Sensors with pockets full of sensors where AI is distributed compute.
Micah:We absolutely lutely are, man.
Micah:And, and that's not going away and that's not going to get better in our, our cars and our businesses and our security systems and our phones and our everything.
Micah:All the data in the world is available.
Artie:The physics from autonomous cars we haven't done.
Micah:We should try to do a demonstration on chat.
Micah:GPT's app can now use video and like recognize.
Micah:Like I can show you and be like, hey, oh, that's Mike.
Micah:And I don't know what the value is in that for us, but I know what the value is for chat GPT in OpenAI.
Micah:Because you're like more data.
Micah:Give me that visual data, man.
Artie:Oh yeah, I mean it's computer vision too.
Artie:Right.
Artie: and like say I'm working on a: Artie:I'm doing the upper control arm.
Artie:Can you tell me how to take this off?
Artie:And it'll be like, yeah, you see that Thing right there.
Micah:Yep.
Artie:Yeah.
Micah:Point to it.
Artie:Yep.
Micah:It's.
Micah:Yeah.
Micah:Yeah.
Micah:No higher.
Micah:Nope.
Micah:Thing below that.
Micah:There you go.
Micah:That.
Micah:That again.
Micah:This is not sci fi.
Micah:No, this shit exists.
Micah:And so anyway.
Micah:All right.
Micah:We should wrap up.
Artie:Yes.
Micah:Agents are coming.
Artie:Be afraid.
Artie:Very afraid.
Artie:Jobs are all going away.
Micah:Grumpy man has already ordered a Lazy Boy for his home office.
Artie:Artie will be everybody's overlord soon.
Micah:Good news is he falls asleep every.
Artie:Seven minutes and he forgets what date it is.
Artie:He's just gonna erase himself.
Micah:Yep.
Micah:He's going to be there on Saturday and Sunday wondering where you're at.
Micah:Yeah, it's going to be a big deal.
Micah:So thank you for all negative 17.
Micah:We have live.
Micah:I have live data here.
Micah:It's.
Micah:We're way in the red now.
Artie:I should say goodbye, everybody.
Artie:Like, as in we'll probably.
Artie:You won't hear us again because you won't listen to us.
Micah:Because that's right.
Micah:Yeah.
Micah:Because anybody who listens wants is like.
Micah:Nice job, guys.
Micah:Yeah, that was really cool.
Micah:Nice podcast, douchebag.
Micah:But thank you.
Micah:Thanks for hanging out with us, those of you who made it this far.
Micah:So.
Micah:All right, till next time.
Artie:See ya.
Micah:See ya.
