Episode 69 – Force Fed a Can of Blueberry Badness

Soundboard: Friends with Brews.

Scott: Brews with Friends?

Peter: I’m more of a Frans with Bros kind of guy.

Scott: But I got questions that need answering.

Scott: Frans with Bros.

Scott: I know you’ve been learning German again because you were confessing your Germanic sins to Adam.

Peter: Yes, actually, and I didn’t do anything today because for the however many days in a row that I’ve been on Rosetta Stone, it has failed to remind me to do my daily practice, even though I have granted it permissions to notify me and told it, yeah, every morning at 8:05 a.m., remind me.

Scott: You also granted it $4,000 or whatever the lifetime membership fee is.

Peter: It was only like, I think it was like 250 bucks or something.

Scott: That’s not bad.

Peter: It was under 300.

Peter: Yeah, no, I decided that that was worth it.

Scott: You should learn more than German, though.

Scott: You should take some other languages on.

Peter: Well, I’ve got all of them there, but I figured that since I have my German citizenship and I would like to go back there, I’d want to make another attempt at learning the language before I do so.

Scott: Mm-hmm, I see.

Peter: So, yeah.

Peter: Woof, hey, what are you drinking?

Scott: Okay, this is weird, man.

Scott: I got myself, I was looking through the store.

Scott: I actually went there to buy some coffee, but I thought, oh, I’ll check out a beer and see if there’s something that I might want to try for today’s episode.

Scott: And I found some, brewed in Korea, but born in New York City.

Scott: I don’t know what that means, beer.

Scott: I think the owner of this beer is a Korean American.

Scott: Anyway, she went back to Korea and she discovered this stuff called makgu, which is a, or makgu, I don’t know.

Scott: I need my wife here to pronounce these.

Scott: Anyway, it’s a rice beer.

Scott: And she started making some alcoholic beverages based on that.

Scott: And this, I think from the picture, looks more like a blueberry yogurt, but this is a blueberry makgu.

Scott: It’s a blueberry rice beer.

Scott: Yeah, well, we’ll just see what it looks like when I open it up here.

Scott: Hold on.

Scott: Oh, wow, it’s got a flavor.

Scott: It’s got a smell.

Peter: Well, okay.

Scott: Oh, I was supposed to shake it because the sediment.

Scott: Damn it, I forgot all about that.

Scott: That’s why the thick yogurt stuff was coming out the bottom.

Peter: Pour it back in.

Scott: So as you can see, it looks a little bit like a smoothie instead of a beer.

Scott: I don’t know if you can tell.

Scott: Do you see that?

Peter: It does.

Peter: Wow, that looks very interesting.

Peter: Right?

Scott: So this could be the worst thing I’ve ever had.

Scott: It is a drink.

Peter: It is a beer, it is a beer.

Scott: I’ve had a blueberry beer on here before, and I really liked it.

Peter: Yep.

Scott: But I’m gonna go out on a limb and say that a blueberry smoothie beer is probably not the best combination.

Peter: Okay, but if you were desperate?

Scott: If I was desperate and somebody said, you have three seconds to get a tiny minor buzz, or we’re gonna shoot your family.

Scott: If they said they were gonna shoot me, I would just say, go ahead.

Scott: If they said they were gonna shoot my family, I would consume one of these for sure.

Peter: Okay, okay, yeah, that’s fair.

Peter: That’s fair.

Peter: I’m sure that the wife and daughter would appreciate that.

Peter: Well, I, dear listener, am drinking a decaf Columbia honey toasted marshmallow.

Peter: And with that, my connection to Scott just dropped.

Peter: FaceTime just aborted.

Peter: That’s great.

Peter: That’s really great.

Peter: I’m going to call Scott back now.

Scott: Oh, now it’s back.

Scott: Yeah, all my devices just went away and then came back.

Peter: So I’ve been recording straight through this whole time.

Scott: That’s fine.

Peter: I was starting to tell our dear listener about what I was drinking, you know, as a way to sort of bridge the gap between this tech hurdle that we are now leaping over.

Peter: The other hurdle is, I can’t actually find-

Scott: Did you say weeping over?

Scott: Cause I am.

Peter: We are weeping over this hurdle.

Peter: Yes, indeed.

Peter: So I can’t actually find this coffee.

Peter: I’m not sure what I’m drinking.

Peter: What I have is Broadcast Coffee, Decaf Columbia, Honey Toasted Marshmallow.

Peter: Yeah.

Peter: I’m not sure.

Peter: I think Broadcast Coffee is the roasters.

Peter: Decaf Columbia, not sure.

Peter: Origin, Honey Toasted Marshmallow, the name of the coffee?

Peter: Or is it Decaf Columbia, the name of the coffee and Honey Toasted Marshmallow is like what I’m supposed to taste?

Peter: I’m not sure what’s going on here.

Scott: Well, show me the bag again.

Scott: Let me find that picture you sent me earlier.

Peter: Yeah, so here’s the thing.

Peter: In my email, they say they sent me Decaf Columbia Palo Rosa.

Peter: And that is what shows up when I did a search for it.

Peter: So that is the right thing.

Peter: But anyway, yeah, so maybe I have Decaf Columbia Palo Rosa, but it’s actually pretty good.

Peter: I’ll tell you that.

Scott: So I did a search on trade for Decaf Columbia and it came up blank.

Peter: Yeah.

Peter: So when I look at it, the packaging is a little different on their website than what they sent me, but it does seem to be like this is it.

Peter: So the description from the website is, there were days when a decaf coffee could never achieve this wonderful balance of brown sugar sweetness and gentle acidity.

Peter: Thankfully, those days are long behind us.

Peter: Tasting notes of nut, citrus, brown sugar, and it’s a medium roast.

Peter: I gotta say, I’m liking this.

Peter: It’s, I think, I’m gonna stay true to my roots.

Peter: I think I would prefer a darker roast, but this is very drinkable, and I’m drinking it without anything in it.

Peter: No contaminants.

Peter: Didn’t put any sugar or cream or anything in it.

Peter: So yeah, this one, this might be an enthusiastic thumbs up.

Peter: Okay, like I would not have to be desperate to drink this.

Peter: This one’s good.

Peter: I’m liking this.

Peter: I’m a little disappointed though, because as I am now back on the slow carb diet, and today is not cheat day, I could not be enjoying a beer with you as we’re recording this.

Peter: And I have a Jacks Abbey rice lager sitting downstairs in the refrigerator.

Peter: So I wanted to have my Japanese style rice beer with you having your Japanese style rice smoothie.

Scott: Korean.

Peter: Oh, oh, so we would be at war with each other.

Peter: Got it, okay.

Scott: That’s why I said, they’re not at war right now, you dumbass.

Peter: Well, not right now.

Scott: That’s why I said it was born in New York City, brewed in Korea.

Peter: Oh, I didn’t catch that part.

Scott: You weren’t listening.

Peter: No, I wasn’t listening.

Scott: It’s okay.

Peter: So, hey, what else is going on?

Peter: You just, you wanted to talk about ChatGPT pricing, and did you notice that ChatGPT just opened up a new thing?

Peter: No?

Peter: You didn’t?

Scott: I wanted to talk about…

Peter: I mean, you put a link in it.

Scott: But that was a sub…

Scott: You skipped over seven things in the…

Peter: Look at the…

Scott: Look below the links.

Scott: Look at the list of topics.

Peter: Look below?

Peter: Well, I was just reading the links.

Scott: Right down here.

Scott: Can you see the stuff I’m highlighting?

Peter: Yeah, I can see that now.

Peter: Don’t blame me for reading top to bottom.

Scott: So when I put these here, all I did was follow on some stuff that you had put here.

Scott: I didn’t start the tradition of putting the stuff down here that we were going to talk about.

Scott: You did.

Peter: Well, why don’t we put the links at the bottom of the page?

Scott: We will, but right now…

Scott: But this is just for us.

Scott: This never…

Scott: I don’t use this when I publish our stuff to the web.

Scott: I don’t use a Google Doc.

Scott: There’s no world in which a Google Doc is wonderful for publishing to the website.

Scott: It just isn’t.

Scott: So I don’t even use this.

Peter: There could be a world.

Peter: We don’t know that.

Peter: There could be a strange new world.

Peter: It could be a world that we would need to be for all mankind.

Peter: You know, there’s a possibility.

Scott: Well, I assume that that world is a world in which people think this tastes good.

Scott: And by the way, when you said enjoying a beer with me, you would have been enjoying a beer alone because I’m not enjoying this.

Scott: This is a thumbs down.

Scott: This is horrible.

Peter: All right.

Peter: So everyone, Scott should let you know that he’s been a little bit moody lately.

Peter: And now I’m not saying that he doesn’t have a reason to be.

Peter: He may have several reasons to be.

Peter: And you know what?

Peter: I may be one of them.

Peter: I may be more than one.

Scott: You’re not one of them.

Peter: But one of the things that Scott is irritated by is AI.

Peter: Not just AI, but many things that are AI adjacent or AI dependent or whatnot.

Peter: And one of the things that I used to irritate Scott, not intentionally, was an AI summary of last night’s Savage Worlds fantasy role-playing game, of which I was a player.

Scott: So here’s the thing.

Scott: That use case irritates me less than the use case of using the AI summaries that it generates for podcast episodes, because those are so off-putting to read.

Scott: They take a podcast and make it sound so boring and sterile and so third-party, so removed from, I know these people, I will listen to these people, that it’s like, why would you…

Scott: Yeah, that would be something that I would never use.

Scott: But, okay, so let’s go on.

Scott: Let’s talk about, first of all, I am curious, when you create these summaries, are these for your own consumption or do you share these with your buddies?

Peter: I share them with my buddies.

Scott: Okay.

Scott: And what do you use them for?

Scott: Do you use them to help you track your progress week to week?

Peter: To remember what did we do last time.

Scott: Okay.

Peter: And also, so here’s the thing.

Peter: When we have ongoing campaigns, I like to keep a record of things.

Peter: Because one of the games I was playing that I was super into and I ran, I don’t remember how many dozens of sessions, was the Superpowers campaign.

Peter: So my amalgamation world of Marvel, DC, Valiant, Image, Dark Horse, plus a heavy dose of Peter’s ideal comic book world all thrown together.

Peter: Astro City, can’t leave them out.

Scott: You didn’t list your employer in that list of heroic organizations.

Peter: Right.

Peter: So anyway, so we kept track of things going along, because there are a lot of story arcs.

Peter: I had like a dozen different story arcs and plot points going along at any one time.

Peter: And even now, I looked back at some of these, this was like four years ago, I had forgotten all like, oh, wow, we did that.

Peter: Oh, that’s right.

Peter: So one of the things that’s tradition in my groups is, like, who can summarize what happened last week?

Scott: Scribe.

Scott: Scribe.

Peter: And so now here’s an example of AI taking away people’s jobs.

Scott: Yes.

Scott: Well, in this case, taking away a job that no one wants to do.

Peter: Well, people do like to, so in Savage Worlds, you get this spendable token that helps you influence the game called a Benny.

Scott: But is it fungible?

Peter: It is non-fungible.

Peter: Well, no, it’s very fungible.

Peter: Anyway, so what we do is we generally give a Benny to whoever does the recap, right?

Peter: So whoever wants to tell us what happened last week, right?

Peter: So that’s where I’ve taken a job away.

Peter: Now, the next thing I thought, and this was inspired by Daniel Measler, and one of his applications of stitching together various patterns was using his tool Fabric to generate the summary.

Peter: So what I did is, first off, I downloaded and installed Fabric, and I only started this like a couple weeks ago.

Peter: And I’ve been playing with it.

Peter: Fabric is pretty cool.

Peter: I think I talked about it on another podcast already, but I’ll touch on it here, because we’re a little more AI-focused here than on the other side of the house.

Peter: So Fabric is essentially a command-line interface to whatever AI model it supports, and easily it supports Llama, it supports Claude, it supports OpenAI.

Peter: And it has a bunch of pre-built prompts, which the author refers to as patterns.

Peter: And these patterns let you essentially, like if you want, for instance, to summarize a document, right, or summarize an article, or extract, analyze arguments, analyze claims made in a document, he’s got all these predefined things.

Peter: So one of the use cases he has is called Summarize.

Peter: So what I did is I wrote another, ran another little utility that he provides called TS, which is Transcribe, or Transcript.

Peter: And I recorded the, so let me back up a step.

Peter: So step one, we do our games, and we use Discord for our audio chat.

Peter: So I used Audio Hijack on my Mac to record the entire conversation in Discord.

Peter: And then I generated a transcript from that using the TS command.

Peter: So it takes everything, generates a transcript.

Scott: What engine does TS use?

Scott: What is it doing behind the scenes?

Scott: Is it feeding it up to a server?

Scott: Is it doing all local processing?

Scott: If so, what model is it using?

Peter: I’d have to double check.

Scott: This is another Measler tool?

Peter: It’s bundled with Fabric, yeah.

Peter: So I think it’s probably using whatever API I set as my default model to do the processing.

Peter: But that’s a good question.

Peter: I don’t even know.

Peter: I think it does it, I’m fairly sure it does it online though.

Peter: It might just be doing a text-to-speech built-in, but I think it’s probably using the Cloud API or the OpenAI API.

Peter: So I use TS to generate the transcript.

Peter: I pipe that into Fabric, and I tell the pattern that I want in Fabric to use is summarize.

Peter: So Fabric then summarizes all of the output that was transcribed.

Peter: From there, this is where it became manual, I copied and pasted it into Cloud, and I said, below is a transcript of a role-playing game with a bunch of friends.

Peter: Tell me the story, summarize it, but retell it in story form, leaving out the stuff about the players and focusing on the characters.

Peter: Because there was one part where the game master was like, guys, I’ve been drinking a lot today, I gotta use the bathroom.

Peter: I’m like, okay, that appeared in the actual summary, but I was like, let’s leave that out of the game, right?

Peter: And then I also gave it a little flavor.

Peter: I said, you know, here are the characters, you know, Helgu, Ulrich, Torgrum and Mittens.

Peter: This is how you spell their names.

Peter: This is their race and their background.

Peter: Just so it knows who these characters are.

Peter: And I’m like, now, tell me the story, you know, or recap the summary in story format.

Peter: So we did all that and it did pretty good.

Peter: And then I took that and went back to 11 Labs and said, generate this now using my voice.

Peter: So now my characters get to hear, or my players, my friends get to hear me, quote unquote me, give them my summary of last week.

Peter: And it’s all very little work.

Peter: The hardest part, honestly, was just getting audio hijack and Discord to stay running and recording the whole time.

Scott: Now you make me regret canceling my 11 Labs account.

Peter: I canceled my own, and then as soon as I did it, I was like, well, I have it for another month, and if I don’t use it, I’ll just recap.

Peter: If I’m not gaining, unless I’m doing an annual type of subscription, I just renew and then I cancel because there’s nothing lost.

Peter: And that way, if I need it again, I just spin it back up again.

Peter: It’s five bucks.

Scott: Yeah, but you lose the voices that you’ve trained.

Scott: That’s the problem.

Peter: Ah, that is a problem.

Scott: So if you have a Peter Nikolaidis voice that you like, you don’t have to-

Peter: I have a Peter Nikolaidis voice that I like, and I have a Scott Wilsey voice that I tolerate.

Peter: And I can reproduce those easily, though.

Peter: I can reproduce those with a drop of an MP3, because we have so many one-sided conversation recordings here that we can generate ourselves either easily anytime we want.

Scott: Okay, number one, it hurts that you say you tolerate it.

Scott: Number two, that’s better than you actually do for me in person.

Scott: So that’s good.

Peter: So that’s progress then.

Scott: Right, it actually is.

Scott: I think that hurts worse than-

Scott: Anyway, where was I going with all this?

Scott: Oh, so I’m looking at Measler’s stuff online, and I got down a rabbit hole before I could find out if he was piping this stuff to the cloud or not.

Scott: Okay, I still have my Peter voice, good.

Scott: So does your Peter voice kind of trail off at the end a little bit unnaturally?

Scott: Like the Peter voice I have for you sounds pretty good, but it’s a little monotone near the end of the sentences in a way that Peter Nikolaidis is not.

Peter: I don’t remember that.

Peter: I do notice, like to me, it sounds blatantly obvious that this is not me.

Peter: It’s almost me, right?

Peter: But then if I am in a certain location, like with my displays, you know, like my main 4K display, my MacBook display, and my iPad all facing me in like an echo chamber configuration, I get some of that same echoey quality that I hear in my AI voice.

Peter: And I’m like, oh, maybe I do really sound like that.

Peter: But since, you know, it sounds different.

Peter: I haven’t listened to my own recording side-by-side just to hear how I sound compared to how it sounds.

Scott: Okay.

Scott: Of the transcripts that you showed me from your game night, the one that I liked the most is when you said, forget about the humans and focus on the characters.

Scott: That one was definitely the best transcript.

Peter: Yes, but again, that was an evolution, right?

Peter: That came from the previous transcripts.

Scott: Yes, yes.

Scott: It’s feeding it through a pipeline of filters in effect.

Peter: Exactly.

Peter: So now what I need to do is I need to write a little, like a command line script to do that automatically.

Peter: And actually, I just had ChatGPT 4.0 write me a Python script to do just that.

Scott: Oh, so I haven’t had a chance to use 4.0 yet.

Scott: In fact, I just noticed today that 4.0 has been introduced recently.

Peter: Like today?

Scott: Yeah, like within the last day or so.

Peter: Yes.

Scott: I don’t know what the benefits are.

Scott: I don’t know anything about it.

Peter: I’m not sure either, but one thing that I did tell it to do was write me the prompt I gave it was write a Python script to process a summary of an audio transcript.

Peter: The transcript is of a role-playing game RPG session.

Peter: Remove references to real life things like players and focus only on the characters, player and non-player, and their actions.

Peter: Present the result in a short story format.

Peter: Prompt the user for names of notable characters, places, artifacts, etc.

Peter: to ensure proper spelling.

Peter: Also, be sure to prompt for the race slash ancestry of a class and class of notable characters.

Peter: And it spat out Python.

Peter: And then I said, please test the code you just gave me.

Peter: And it ran the Python analyzer and it spat out some output.

Peter: So theoretically, I have some working Python code.

Scott: So, you had it make a Python script so that it could do each of these steps sequentially?

Peter: No, just the last step.

Peter: Convert summary to fantasy would be this last step.

Scott: But why do you need a Python script as opposed to just a prompt for that?

Peter: Because I want to do it on the command line as part of a chain of things.

Peter: I don’t want to have to copy and paste it into a UI.

Peter: Yeah, so I want to automate the whole thing so that I’ll be able to say, I’ll just write a little script called Summarize Game, and it will do the transcript, it’ll do the summary, and then it’ll do the conversion of the summary to story format.

Scott: Have you ever thought about the fact that you could write a small script called Peter Nikolaidis, and then you wouldn’t have to hang out with people that have been drinking too much and need to go off and drink water?

Peter: That’s the point.

Peter: I don’t have to be around to tell them about the summary.

Scott: No, no, you don’t even have to be there to play the game.

Peter: Well, we’re getting there.

Peter: Soon, it’ll be just a bunch of AIs playing the game, and the games tell us what happened.

Scott: Exactly.

Scott: Your Peter agent will play your game for you.

Peter: Yep.

Scott: So, anyway.

Peter: Yeah.

Peter: Oh, can’t wait.

Peter: So, that’s cool.

Scott: Here’s the problem with the main summarizing capabilities.

Scott: If you just say, take this and summarize it without all the additional work that you did.

Scott: And this is true of the podcast summaries that I’ve read generated by AI as well, including the one that you sent me.

Scott: Yep.

Scott: The AI is to podcast and other conversation transcripts as George Lucas is to Star Wars.

Scott: They know there’s something interesting there.

Scott: They have no idea what it is or why people like or don’t like it.

Scott: And so, all kinds of ridiculous things are included that you just don’t care about as a reader.

Scott: And some things that you might care more about are just glossed over in the most banal terms possible.

Scott: It’s like beating the excitement out of that dying horse.

Peter: But that’s, yes, and that’s exactly why I gave it the further on, you know, prompt down the road.

Peter: I’m like, forget about the fact that my friend had to go pee halfway during the game, right?

Peter: I don’t care about that.

Peter: Don’t tell, you know, whenever you hear John, because John hasn’t entered his character’s name in our game, so he shows up as John, actually say, Oolag, you know, because that’s John’s character, right, et cetera, et cetera.

Peter: It’s an iterative process.

Peter: It’s, you know, iterative approach, we’re getting better each little step of the way.

Scott: Yeah, I just find that AI writes in a way that it’s very off-putting.

Scott: I don’t know how to describe it other than to say it is boring as utter hell.

Peter: So, I could try, say, you know, writing this summary in the style of George RR.

Peter: Martin.

Scott: Or Douglas Adams, or?

Peter: Sure.

Peter: That would be a little more appropriate to this crew, though.

Scott: Well, I know there’s somebody named Mittens.

Peter: Yes, well, that’s, yes.

Peter: So, so Mittens, and there’s the thing is, like, the game didn’t know, the game assumed that Mittens was a her, right?

Peter: Well, my friend who plays Mittens hasn’t specified, but he’s a he, and he generally plays he’s.

Peter: So I told him Mittens is a male Rakshaasan, you know, and Rakshaasa is a cat type, you know, cat type humanoid.

Peter: So, you know, I clarified that, give it a little bit.

Peter: I let him know that Torgrim is my character, you know, my half giant character, for example, gives it a little flavor.

Scott: Okay, now I’m not a gaming role player.

Scott: But when it says things like so and so did something, but inadvertently or failed, I don’t remember what the adjectives that used were, but it used the same one every time, but it failed massively or whatever it said.

Peter: Yes.

Scott: And then something happens.

Scott: How does that get determined?

Scott: Is that determined by you guys coming up with that?

Peter: So the rolls of the dice determine that something either succeeds, fails, epic succeeds, or critically fails.

Scott: And then you can embellish on that.

Peter: And then you can embellish.

Peter: Exactly.

Peter: And that’s a combination of like, you know, the game master has the final say.

Peter: So in my character, for instance, my character Torgrim, you know, he’s a giant.

Peter: He’s a half giant, so he’s 10 feet tall.

Peter: And any time he’s in like normal size environments or even worse, in this case, dwarven sized environments, it’s hard.

Peter: So he’s crawling through these tunnels on hands and knees while his friends are, you know, some of them are crouching a little bit and others are just walking right through.

Peter: We get set upon by all these guys and he’s trying to fight.

Peter: He also has this hindrance in the game called blunderer so that any time normally in the game, you get to roll two dice.

Peter: It’s a long story.

Peter: But anyway, he has this hindrance where more frequently than others, he will blunder and do something epically bad.

Peter: So he was in hand-to-hand combat alongside with the knight, and he reached to try to grab one of these giant centipedes and critically failed.

Peter: And I was like, well, I envisioned that he grabbed the knight’s shield and just ripped it out of his hand by accident instead.

Peter: And you know, the player who I negatively affected, he was like, yeah, that tracks, so there was no objection, so that’s what happened.

Peter: You know, someone who could have been more lenient could have just said like, oh yeah, you missed and stubbed your toe.

Peter: No, no game mechanics.

Peter: Someone else who might be a little more, a little less lenient, a little more harsh could say like, oh no, you actually missed and you grabbed the knight’s sword by accident and inflicted a wound on yourself, right?

Peter: So that’s where the game, you know, there’s mechanics, there’s opening, open for interpretation in the rulings.

Peter: And that depends, you know, that depends on the setting that you’re playing.

Peter: It mostly, however, depends on the people, you know, the crew that you’re playing with.

Peter: So I will often, I find myself, I’ll give the players the opportunity.

Peter: I’m like, so tell me, you know, like, what does that look like, right?

Peter: Or, you know, how does that play out?

Peter: And, you know, if they’re like, oh, you know, I dropped my sword.

Peter: I’m like, OK, critical failure is supposed to be the worst possible outcome.

Peter: I think you could probably do a little better than that, right?

Peter: You know, so nudge them a little bit and like, OK, you dropped your sword, or maybe, you know, something else happened as well, something like that.

Peter: So there you go.

Scott: So here’s the type of thing, like when you sent me the summary about FWB, that’s Friends with Brews for those of you listening along.

Scott: Here’s the thing that I find completely useless.

Scott: The conversation appears to be a casual discussion between two individuals, likely podcasters, about various topics, including TV shows, software and personal experience.

Scott: The conversation is informal and often humorous with the speakers using colloquial, you know what I’m trying to say.

Peter: Colloquial.

Scott: Colloquialisms.

Scott: Thank you.

Scott: Geez, I do know how to say that word and sarcastic remarks.

Scott: Like, I don’t, I wouldn’t, why would you put, I would never.

Scott: I don’t get it.

Peter: Right.

Peter: I mean, it’s very sterile.

Scott: Yeah.

Peter: It is a, yeah, I get that.

Scott: It’s like, I must understand how humans interact.

Peter: Right.

Scott: So, I guess what I’m saying is, it has, I don’t know why people worry about AI taking their jobs.

Scott: I think those people haven’t actually watched and seen the output that AI gives in most cases.

Scott: Because AI is useful for a lot of things.

Scott: But, and here’s the, here’s the intriguing thing.

Scott: People, creatives are worried that AI is going to take their jobs.

Scott: And in my opinion, very unnecessarily so at this point.

Scott: Right now, it is not equipped to do anyone’s job in terms of arts, in terms of writing, in terms of what you would call, you know, humanities type skills.

Scott: AI has a very bad reputation when it comes to programming, which I don’t understand anymore.

Scott: I think it’s gotten quite good.

Scott: At least ChatGPT 4 and higher have gotten quite good at programming, or returning programming answers, I should say, more correctly.

Scott: I think that people have seen how it has been in the past, or probably how it is with specific languages that I don’t use, maybe, let’s say.

Scott: I’m sure there are some cases where it falls flat on its face, no doubt.

Scott: But overall, it’s pretty freaking good and useful.

Scott: And there, I think people are relying on impressions from the past.

Scott: And then when it comes to the humanity stuff, I think people are afraid of something that hasn’t occurred yet, and we don’t know if it ever will.

Scott: That is the point to which it’s so good that I want to consume it.

Scott: Like, if you read the AI-generated trash on the web that passes for news that all these sites are putting out, it’s horrible.

Scott: It’s literally just content.

Scott: And that is the only thing you can say about it is it’s a place to slap an ad.

Peter: Yep.

Peter: Yeah, totally.

Peter: And you can tell.

Peter: You sent me, I think, the article, the announcement of…

Peter: What was it?

Peter: Was it a Microsoft announcement the other day or something?

Peter: There was something about AI.

Peter: Was it Copilot?

Peter: I don’t even remember.

Peter: There was going to be…

Peter: Or maybe it was talking about OpenAI and, you know, chat GPT-5 or something like that.

Peter: It was talking about Microsoft and how Microsoft, you know, when they bought Anthropic, so maybe they’ll use Claude.

Peter: And I was like, whoa, no, that did not happen.

Peter: But I did just hear today, which I don’t know how I missed this, that Microsoft bought Inflection.

Scott: Yeah, right.

Peter: I did not catch that.

Peter: I was like, how did I miss that one?

Peter: But, you know, the hero was, you sent me an article, which seemed like a good article, but I get this sneaking feeling that that was AI generated, or, you know, it could have just been a, you know, clumsy human who didn’t know what they were talking about too.

Scott: It could have been a 10-foot tall guy scrambling around in a cave trying to fight people.

Peter: Crawling on hands and knees.

Peter: Yeah, so there you go.

Scott: Yeah, I don’t know.

Scott: Anyway, real quick, let’s start going through it.

Scott: One thing I wanted to find out is what stuff you’re still using.

Scott: So I got kind of an idea there.

Scott: I want to know how much of, as we talk about this, I want to know how much of this stuff you’re doing on-device, because, like, you were compiling llama or something like that, right?

Scott: So how much of your disk space are you giving up to large language models?

Peter: OK, so right now, I’ve given up a little over 40 gigs to llama.

Peter: OK, now, here’s the thing.

Peter: I signed up for my…

Peter: I was like, OK, I’ve heard enough people talking about llama.

Peter: It’s time to do it.

Peter: I mean, you know, like, I’m spending, what, maybe a few bucks a month on, you know, chat, GBT and Claude APIs.

Scott: What size SSD do you have in your Mac?

Peter: Terabyte.

Peter: You know, it’s not costing me an arm or a leg, but let me try this local thing.

Peter: Let me see what this is all about.

Peter: So I downloaded Llama 3.

Peter: I went through their use case, downloaded this huge file, and, like, the download failed.

Peter: And I had to work around that.

Peter: I got the download to work.

Peter: Okay, fine, now I have to build it.

Peter: The build failed.

Peter: Then once I could get the build working, it wouldn’t actually execute.

Peter: And I’m like, okay, and I was just, you know, I get frustrated.

Peter: I’d give up, and then a week later, I would try again from scratch.

Peter: And then I found Olama.

Peter: And Olama gives you a binary pre-compiled for your operating system, Mac OS in my case.

Peter: And it just downloads and does the install for you.

Peter: And suddenly I just like run, you know, curl such and such this install script, blah, blah, blah.

Peter: And suddenly I have Olama.

Peter: I have Olama 3 installed on my Mac.

Peter: So I’m like, whoa, sweet.

Peter: And then I was able to, once I had that running, I was able to point fabric at that, or I could just run Olama from the command prompt and interact with it just like I would via ChatGPT, like via the web interface.

Peter: So I can do the same thing.

Peter: And for a quick little snippet, you know, like ask me, like, you know, generate a non-player character, you know, generate an NPC cabbie for my superpowers campaign or something.

Peter: It’s okay.

Peter: It was a little slow, right?

Peter: Now again, I’m running a MacBook Air M2, right?

Peter: So not state of the art.

Peter: But I was like, hey, it’s free and it’s private.

Peter: That’s the big thing, too.

Peter: It’s good, because I want to work on some client documents.

Peter: I’m really curious.

Peter: I have been going through and spending hours upon hours updating one of my clients’ policy set to bring them up to speed from PCI version 3.1 and 3.2 to version 4.0, which is the current, you know, PCI DSS.

Peter: And I’d be curious just to say like, here, here’s a policy document.

Peter: Tell me, you know, what references to PCI version 3 need to be updated to 4 and how to make those changes.

Peter: I just want to know, can it do this, right?

Peter: But I don’t want to take my client’s proprietary documents and just throw them out to some LLM, right?

Peter: So working locally with that, I might be able to do this.

Peter: And this was inspired by the…

Peter: A couple weeks ago, I watched the SANS AI conference, and it was pretty good.

Peter: It had a few interesting takeaways.

Peter: So anyway, it got me reinvigorated, and now that I had a week off from work because I’m just sitting around recuperating, letting my shoulder heal up, I was able to actually make some significant progress.

Peter: That said, I also have a little old Intel Nook-style device here running Linux, because sometimes, despite all the similarities, there’s stuff that might be working or precompiled for Debian, Linux that doesn’t exist for Mac.

Peter: So I was like, let me just try it on this box.

Peter: It’s so underpowered, it’s not funny.

Peter: So I’ve done a little bit of local processing, but not a ton.

Peter: And I just started playing with it right now.

Peter: But I tried, for instance, to do those transcripts that I showed you last night, and I ran overnight.

Peter: There were three files, and each one got longer.

Peter: The first one was 20 minutes long, the next one was an hour, and then the final one was an hour and a half.

Peter: And I ran it overnight on my M2 against the Llama 3 model, and it finished the first summary.

Peter: I ran them again this morning against Claude, and it was done in five minutes.

Peter: Because they’ve got a lot more power there throwing at it on their servers than I am here.

Scott: So I just found an article saying, running a Llama on the Raspberry Pi.

Scott: So basically what you’re telling me based on your performance with an M2, I don’t think you would want to run a Llama on a Raspberry Pi.

Peter: I do not see anything running on a Raspberry Pi.

Peter: I mean, I see that Llama being…

Scott: Llama 3 runs on a Raspberry Pi.

Peter: That Llama is getting dragged across the mountains.

Peter: It is not running anywhere.

Peter: Yeah, so anyway, I have a Llama, and this gets back to a discussion you had with Ronnie.

Peter: I didn’t even realize, because I’ve been working downstairs for the last several days sitting on the couch or at the breakfast bar with my laptop, I brought my laptop upstairs, plugged it into my external display, and I saw that, hey, I have a little finder bar, menu bar icon for Ollama.

Peter: Oh, and Backblaze.

Peter: Oh, and NordVPN.

Peter: Look at all these things I didn’t realize had icons for, because…

Scott: Wow, did they just update today or something?

Peter: I don’t know, but here they are.

Peter: So helpful, Apple, for just making all these things vanish from sight when you have a MacBook.

Scott: I know, I just don’t get it.

Scott: Just making them disappear without a single word is just like, you won’t see these.

Scott: Good luck.

Scott: It just is so baffling to me.

Scott: This is from the company that’s supposed to be the king of simple user interfaces.

Peter: You are, man.

Peter: Yeah, and the funny thing, too, is I’ve spent…

Peter: I’ve known when I’ve wanted to find a certain icon, right, up on there, I would start closing things.

Peter: Like, I close my DynDns client.

Peter: I close Shotter.

Peter: I close Rectangle Pro.

Peter: I sign out of OneDrive.

Peter: You know, I’m closing all these things, hoping just to find this one icon.

Peter: Oh, there it is.

Peter: So…

Scott: You should never have to…

Scott: That is insanity.

Scott: It’s crazy.

Peter: Mm-hmm.

Scott: All right, so here’s what I’m doing right now.

Scott: I am still subscribed to ChatGPT Plus, because I find it useful-ish, do I?

Scott: I think I could probably give up my ChatGPT Plus, subscription.

Scott: Mm-hmm.

Scott: I have OpenAI API, which I will keep, because I have several apps that can use an API key.

Scott: And it’s kind of handy and useful to be able to do that.

Peter: Right.

Peter: Which is separate.

Peter: The listeners should know this, but if they don’t, your API access and your monthly pro subscription access and Claude does this the same way.

Peter: They charge separately for those.

Peter: One’s a monthly subscription, and the other’s a per use or per token type of thing.

Scott: Yeah, and a lot of apps will say, okay, enter your API key, and then they’ll use whatever models are available to you through your API in the app.

Scott: And that’s what the API is useful for.

Scott: Among other things, you could write your own apps against the API or your own scripts or whatever.

Scott: Then I have the Raycast AI add-on, which I like because it’s just so handy to invoke, and I just like having it there all the time.

Scott: I also have the Notion AI add-on, and I am going to…

Scott: I say we’ll almost certainly get rid of this one.

Scott: No, I’m absolutely going to, because I’ve literally never used it.

Scott: And…

Peter: Which one?

Scott: Notion AI.

Peter: Notion AI, yes.

Peter: Yeah, I used it by accident once or twice.

Scott: Right, you have…

Scott: Yeah, you get a very limited version built in, but you can pay more to get the AI add-on.

Scott: I paid for that for some reason.

Peter: I’m thinking that Notion AI…

Peter: Like, if you…

Peter: I want to say, if you work at Notion, or if you are one of these hipsters who is…

Peter: You’ve designed your whole life around Notion, which you could do, right?

Peter: You could use Notion.

Peter: I could use that…

Peter: No, I wouldn’t say I can replace Excel with it, but I could certainly replace Word with it for a lot of the stuff I do.

Scott: People publish stuff to the web through Notion.

Peter: Oh, absolutely.

Peter: No, absolutely.

Peter: People blog with it.

Peter: People have their documentation wikis in it, et cetera, et cetera.

Peter: You could definitely do that.

Peter: I’m not saying you should, and I probably won’t.

Peter: But if I was in Notion, like, you know, 50% of the time, I would benefit from having Notion AI.

Peter: Just like you, you know, you’ve really embraced Raycast.

Peter: I use it as a glorified task switcher.

Peter: You know, that’s really still what I do most of the time.

Peter: That and window resizing, that’s mostly what I do.

Peter: But if you spend a lot of time in Raycast AI, that’s great.

Peter: The one application-based AI that I’ve been using the most is Warp, Warp AI.

Peter: And still just the free version.

Peter: I haven’t bumped up against their limit just yet.

Scott: Their limit’s pretty low.

Scott: It’s 40 requests per month, but like you, I haven’t bumped up against it yet.

Scott: And once you told me that you were using it, I started doing use cases with it.

Scott: And some of the nice things are being able to select something in the terminal and say, explain this or do something with this.

Peter: Bingo.

Peter: Or, you know, like, I just typed this command.

Peter: How do I…

Peter: I can never remember the, you know, the PR command or, you know, or whatever.

Peter: Like, how do I get just the third column from this and then uninstall anything that shows up?

Peter: It’s like, oh, you want to do this?

Peter: I’m like, yes, perfect.

Peter: But I did notice that the Warp AI is tailored to, like, command line use cases.

Scott: Yeah.

Peter: You can’t ask it a general question about, like, you know, like, what was the main campaign point of, you know, Barack Obama in 2000 or something?

Scott: Yeah, yeah, yeah.

Scott: No, it’s very much geared towards, you’re in the terminal.

Scott: What else would you be asking it about but things that are in the terminal?

Scott: Which is fine with me, yeah.

Scott: It’s focused, and that’s what it needs to be.

Peter: I still, I feel like I should, I have not still even tried, like, Microsoft Copilot anything.

Peter: I’m really curious to see if it’s any good.

Scott: I really like Copilot, and there’s a replacement Copilot that people keep talking about, and I’m, damn it, I just closed a tab I wanted.

Peter: Well, it’s not OpenPilot, because that’s what drives my car.

Scott: It is called, oh, Codium, I think, is the one.

Peter: Codium, don’t know that one.

Scott: Codium, yes, Codium’s the one.

Scott: And I’ve tried it.

Scott: It’s okay.

Scott: But it uses older models more often than not.

Scott: And it’s just not, I don’t know, in my use case, it’s not as integrated in VS Code.

Scott: I don’t know, it’s like, I need to use it again.

Scott: I know they’re improving it all the time.

Scott: It seems like it has a lot of work to go before it actually defeats Copilot for me.

Scott: But Microsoft Copilot in terms of programming, for example, with like VS Code or something, it’s pretty integrated and it works really well in my experience.

Scott: It does a lot of nice things like auto-completion of code.

Scott: You start typing something, it’ll show you, hey, here’s what I would do if I were you.

Scott: And sometimes it’s great.

Scott: Sometimes it’s perfect.

Scott: It’ll give you an entire function and it’ll do exactly what you want.

Scott: Or you just need to edit a couple variable names or something like that and it’s perfect.

Scott: And so for that kind of thing, it’s nice.

Scott: It’s also nice to say, hey, in the context of this.

Scott: Now, supposedly it’s also good at knowing what other pages you have open, what other files you have open in VS Code and referring to those.

Scott: I haven’t had a lot of experience where I needed to have it do that.

Scott: In my experience, I haven’t really seen that yet.

Scott: But anyway, it’s pretty good, but of course it is another $100 per year or something like that.

Scott: So it’s got to be, I don’t know.

Scott: All these things are…

Scott: That’s the problem.

Scott: All of these things have their uses, and they all cost a separate fee every month.

Peter: Yep, and that’s just it.

Peter: So you got to be selective, right?

Peter: If you’re a Daniel Measler or a Jaden Schafer and you’re building your life around AI, then what’s another $5, $10, $20 a month?

Peter: Whatever, it’s worth it.

Peter: And I’m trying to myself…

Peter: I’m getting a little less cheap recently, thanks to Adam Bell reminding me, like, dude, you’re being cheap.

Peter: Although he never says that.

Peter: He just retroactively pointed out when I was being cheap.

Peter: But anyway, thank you.

Peter: Thank you.

Peter: Just the same, Adam.

Peter: But that said, I still review my subscriptions every now and then, but I will…

Peter: I’ll add it.

Peter: Let’s spend $5 a month here.

Peter: Let’s spend $10 a month.

Peter: Let’s see if it’s any good.

Peter: And if I don’t like it, then after a month or so, I just cancel it.

Peter: And if it’s bringing some value, then great, you know?

Peter: It doesn’t take a lot.

Peter: It doesn’t take a lot for me to get $5 worth of value for something.

Peter: That’s a small lift.

Scott: So there you go.

Scott: Yeah, I think…

Scott: I don’t know.

Scott: It’ll be interesting to see where all this shakes out.

Scott: I do think that…

Scott: I think people should be careful, because there is a knee-jerk reflex against AI that I understand.

Scott: But like I said, with the programming thing, I think people are badmouthing it where it’s actually gotten quite good at it.

Scott: I also think that a lot of people are fearing AI for their jobs where they don’t necessarily need to, or at least not yet.

Scott: And I could be short-sighted.

Scott: Hey, I’ll be the first to admit, maybe I’m wrong.

Scott: Maybe I don’t see the cascading wave of failure headed towards these people, but I wouldn’t worry about it right now.

Scott: I think there’s a little bit too much angst.

Scott: But I think the things we should be worrying about are just how much energy consumption this stuff uses.

Scott: That, to me, is the biggest issue with AI.

Scott: It’s none of these other things.

Scott: The problem is we’re already living in a world where we’ve got some serious ecological problems going on, and we’re just going to compound those by piling on the compute power, right?

Scott: Everything needs compute power.

Scott: Everything needs massive compute power.

Scott: In the world that we live in right now, that requires tremendous energy.

Peter: So you’re saying that humans are wasteful?

Scott: I am saying that humans…

Peter: Inefficient?

Scott: Are inefficient and do use a lot of resources to do things that ultimately don’t matter.

Scott: Obviously for me, stuff like Bitcoin, crypto, that is the biggest waste of natural resources that you could imagine.

Scott: It’s just a scam.

Scott: We’re destroying the planet for a gigantic Ponzi scheme, in essence.

Scott: AI, I think, definitely has more utility than that, but it also, at the same time, is definitely a fad.

Scott: Every company has to AI all the things.

Peter: Yes.

Peter: Everyone is AI-ing everything these days.

Peter: And yes, I mean, it’s like…

Peter: Did you send…

Peter: I forgot if I sent it to you, the picture of every company with AI, and it was the fast food…

Peter: We should put a link to that if we can find that.

Peter: That was just classic.

Peter: Just throw in an extra sauce on everything, you know?

Peter: And it’s like even on the bag.

Scott: Yeah, on the bag, on the outside.

Scott: Yeah, everything has AI.

Scott: Yeah, and that leads to wasted resources, and probably wasted human resources too at work, because everybody’s focused on something that ultimately their use case probably won’t provide that much value to humankind, you know what I mean?

Scott: Like a lot of these things that we’re throwing AI at aren’t really going to benefit us at all, and all they’re doing is wasting tons of electricity and water for data centers.

Scott: So, yeah, you know, AI has promise.

Scott: It definitely has uses, but right now, we’re throwing everything at it, including the planet.

Scott: And I feel like I wish humans could do a better job, but this is the problem with capitalism, and I’m not going to go on an anarchist rant here, because I hate anarchy.

Scott: Capitalism says you have to make money and you have to take advantage of what’s hot.

Scott: And it doesn’t allow for successful businesses that aren’t constantly growing.

Scott: Everything has to be growing, and this is why we destroy the planet to follow these fads, is because everyone has to get in on it, everyone has to make their millions.

Scott: Instead of just saying, what is this actually really good for?

Scott: Let’s take our time and figure it out.

Scott: That will never happen.

Scott: We don’t have that kind of introspection, we don’t have that kind of deliberation.

Scott: I don’t know, that part of it is depressing to me, because it just shows…

Scott: And I think we’ve gotten worse about jumping on bandwagons in recent years, computing bandwagons anyway.

Scott: Do you think that’s true, or do you think I’m just becoming more tired of it?

Scott: Because in the old days, I feel like the stuff that we jumped on wasn’t as immediately harmful to society or the planet.

Peter: I think it does feel like it’s escalating a little bit.

Scott: Maybe we’re just old.

Scott: I mean, I know we’re old.

Scott: Could be.

Scott: I don’t think it’s a maybe, I think it’s a given.

Peter: But is that the problem?

Scott: Right, yeah, are we the problem?

Peter: So, I have a related thing.

Peter: So, in addition to along the lines of using things that we’re not using anymore or whatnot, so I still have that S-GPT menu bar item, and I have not used S-GPT in months.

Scott: I don’t even know what that is.

Peter: You turned me on to it.

Peter: It’s basically a menu bar interface to ChatGPT.

Peter: It was the first one that we used outside of the regular chat.

Scott: I thought the first one I used was Mac GPT.

Scott: Oh, S-GPT is the shortcut.

Scott: No, S-GPT is Federico’s shortcut.

Peter: Oh, OK.

Scott: Are you sure you’re not thinking of Mac GPT?

Peter: No, definitely this is S-GPT, and it’s in my menu bar, and I don’t know how to get rid of it, because when I search my file system for it, the only option I have is to launch it, and I don’t know where it came from.

Scott: So open shortcuts.

Peter: So the shortcuts can put themselves in your menu bar?

Peter: Oh, yeah, look at that, menu bar right there.

Peter: There it is.

Peter: Look at that.

Peter: Remove from menu bar.

Peter: Okay, done.

Peter: Perfect.

Peter: Thank you.

Peter: That’s a shortcut.

Peter: I should have known.

Scott: No problem.

Peter: I was going to say, because I was about to say, I was like, isn’t S-GPT the thing I had in shortcuts on my iPhone?

Peter: What’s it doing up there?

Scott: Right.

Scott: Yeah, I don’t use S-GPT anymore either.

Scott: I think that, going back to the transcript thing, please do not at all think that I am poo-pooing the idea of you generating, not transcripts, but oh, I did want to talk about transcripts very briefly.

Scott: I don’t poo-poo the idea of generating summaries at all.

Scott: It’s the particular output that it gives in some cases that I find very annoying.

Scott: Not the fact that Peter Nikolaidis says I should generate summaries.

Scott: Yes, you absolutely should.

Scott: And what better tool to try to do it with?

Scott: And I think the fact that you’re working through it and getting it towards a better summary is really cool.

Scott: I like your experiments with this.

Peter: But when you pointed out how annoying those summaries were, that was what spurred me to say, well, how can I make this better?

Peter: And so, like I said, you just fed into the iterative process there, so it’s all good.

Scott: One thing I want to say about transcripts is, how did you generate the transcript for Foubois?

Scott: What did you use for that?

Peter: That was also the TS command.

Peter: Okay.

Scott: So I was using Mac Whisper.

Scott: Was it Mac Whisper?

Scott: Is that what I’m thinking of?

Scott: I was using Mac Whisper to generate transcripts, and it’s pretty good.

Scott: It’s not bad.

Scott: It really is pretty good.

Scott: However, Apple Podcasts, whatever they’re doing on their servers to generate transcripts, it’s second to none.

Scott: It is so bloody good.

Scott: So now what I do is I just log in.

Scott: I wait a few days.

Scott: I log in to Podcast Connect, and I download the transcript that they generate for the episodes.

Scott: And then I just clean it up a little bit with some formatting and stuff, and I publish those.

Scott: And those are super, super good.

Scott: And they finally caught up to our backlog.

Scott: So now I have transcripts for all episodes on my hard drive.

Scott: I don’t know why I call it a hard drive.

Scott: It’s an SSD, Peter.

Scott: I have them.

Scott: I will get them published pretty soon.

Scott: So now there won’t be a big gaping hole between episodes 14 and 50, whatever it was anyway.

Scott: I’ll get all those published and all episodes will have transcripts.

Scott: They’re not all in exactly the same format because I kept changing the way I was doing them.

Scott: And then when I switched from MacWhisper to Apple Podcasts transcripts, I definitely changed the way I was publishing them.

Scott: So forgive the inconsistency.

Scott: At some point, I may try to clean them all up and make them consistent.

Scott: But the fact is that if you have a podcast and you put it in Apple Podcasts, you don’t have to do any work to get transcripts.

Scott: Apple is happily doing that for you, and they’re doing a very good job of it.

Peter: So there you go.

Peter: Now, if we start podcasting our Sunday Night Games, then we’ll be in good shape.

Scott: Do you know what else seems like it was generated by AI?

Scott: This beer, that beer that I drank, seems like it was generated by AI.

Scott: I don’t think any humans were involved, because it is not good.

Scott: Oh my God.

Scott: It is like a smoothie with alcohol.

Scott: And I’m sorry, I like smoothies.

Scott: I like some alcohols.

Scott: I don’t like smoothies and alcohol.

Scott: It was not good.

Scott: Now, I have a Norm.

Scott: I also bought some others of theirs that is just the plain, the original plain rice beer.

Scott: And I’m hoping that it’s a clear fluid, not a smoothie-ish thing, and I’m hoping that it tastes better.

Scott: But this blueberry thing was bad.

Scott: And now I want to have a confession to make, a pre-confession.

Scott: This is crime thought.

Scott: Whenever you determine from my brain that I’m about to commit a crime in advance, I’m going to give one of those blueberry things to my wife and see what she thinks of it.

Scott: She might like it, though.

Scott: I might not be torturing her.

Scott: And trust me, if she doesn’t like it, she’s an intelligent whole human in her own right.

Scott: She’s going to stop drinking it after one sip if she doesn’t like it.

Scott: It’s not like I’m going to force feed her a can of blueberry badness.

Peter: Excellent.

Peter: Well, dear listener, on that note…

Scott: Maybe I should do that as the title, force fed a can of blueberry badness.

Peter: Force fed a can of blueberry…

Peter: There’s your title right there.

Peter: Dear listener, if you want to find us, you can find us at friendswithbrews.com.

Peter: You can find Scott there.

Peter: You can find me there.

Peter: You can find us both there.

Peter: And with that, I think it’s time to put a pin in it and push the big red button.

Soundboard: We’re done.

Soundboard: Remember?

Soundboard: This is what we agreed to.

Soundboard: The last step.

Soundboard: We stick with the plan.

Soundboard: We leave.

Soundboard: Tell your friends.