Episode 005 · The AI Zeitgeist
Muse, Jev, and the AI That's Always On Exploring Meta's Muse and TypeSafe's Jev.
This week, Austin and Luke dive into Meta's new AI agent, Muse, exploring its capabilities and quirks. They also discuss Jev, a groundbreaking AI model from TypeSafe that's changing the game with its low-cost, high-speed processing.
▶ Now playing
Muse, Jev, and the AI That's Always On
0:00 / 20:01
−15s +15s 1×
0:00 Cold open 0:26 Introduction to Muse 1:11 Using Muse for Tasks 9:22 AI's Accelerating Development 12:22 Introduction to Jev 14:52 Jev's Applications and Implications
Pull quotes “It's just the fascinating part that it does things for you and it remembers what you've been working on.”
Luke · Play from 1:00 “It definitely does seem more autonomous than the other AIs.”
Austin · Play from 5:56 “The most important thing about AI is to know how to use it incredibly well but then also touch a lot of grass.”
Luke · Play from 10:31 “Jev can process these type questions extremely fast and extremely cheaply.”
Austin · Play from 15:39 Full transcript 0:22 Austin: Hey welcome to the AI Zeitgeist0:24 Austin: I'm Austin0:25 Luke: And I'm Luke0:26 Luke: And this week I wanted to share my experiences using Muse0:31 Luke: It's a new AI agent that is by Meta0:36 Luke: And I was actually on Facebook I was and saw an for it and I was like oh I've0:40 Luke: heard of this0:40 Luke: I heard that Meta had something called Muse and I to try it out and see what it0:44 Luke: was like0:45 Luke: Downloaded it got it started jumping into it0:48 Luke: And it's very interesting because the fascinating part of it is not necessarily how good or0:56 Luke: how bad the model is what the benchmarks are0:59 Luke: I'm not even thinking about any of those things1:01 Luke: It's just the that it does things for you and it remembers what you've been working1:06 Luke: on previously and then it will try to take actions on your behalf but in a permissioned1:12 Luke: way1:12 Luke: It's not going to just go out and do stuff1:14 Luke: It's going to say like I can do this for you or I can do that for you1:17 Luke: And then you have to say like yes I want you to do that before it actually does something1:21 Luke: But the cool thing about it is that it initiative1:25 Luke: So you will be thinking of something maybe you've chatted with it you get in some context1:30 Luke: it knows what you're1:31 Luke: I was looking at possibly getting something off Facebook Marketplace and I was like can1:35 Luke: you go and the lowest price on this thing and what is the way to do it1:41 Luke: And it was like it found a of things and it's like Facebook Marketplace1:45 Luke: for me1:47 Austin: And1:47 Luke: apparently it will I can set a buy alert for like something that's like a1:54 Luke: condition like good or like new1:57 Luke: Right1:57 Austin: So it has to be at this condition1:59 Luke: Exactly2:00 Luke: And then it's like and like this price or lower2:03 Austin: Right2:03 Luke: And then it will like with the2:06 Luke: And it will set up a meeting time and place and2:09 Austin: I2:10 Luke: just show up and pay the cash2:12 Austin: Right2:12 Luke: I haven't used it yet in that way2:15 Austin: That's interesting2:16 Austin: And I if like the sellers can have muse2:19 Austin: So it just ends up being two AIs negotiating the price of this Lego set from 19902:26 Luke: Because if you know what2:27 Austin: your2:28 Luke: point is and you know what you would agree to or not agree2:32 Luke: to and you give that to the agent then it can just take care of that for2:35 Austin: you2:35 Austin: Right2:36 Austin: Right2:37 Austin: Yeah that is pretty cool2:38 Austin: So I actually just downloaded it this past week I think to use it2:44 Austin: And it's pretty interesting2:46 Austin: And the first thing I had to do because you're always like when you get into a new thing2:49 Austin: you're like what do I do2:51 Austin: What do I ask it to do2:52 Austin: Which is a legitimate2:54 Austin: That's one thing that people struggle with is you have an empty chat input and you're3:00 Austin: like what do I put in this3:01 Austin: What do have to do3:02 Austin: So I had to clean up my email my Gmail account and it did an OK job3:07 Austin: But well actually I had Grokbot do that and it didn't do an OK job3:11 Austin: So then I had Muse do it and it did a pretty good job3:14 Austin: And then now I've like it's3:17 Austin: It will send me notifications like push notifications on my phone3:22 Austin: Like it said hey it looks like you it looks like a secret key was created on GitHub the3:29 Austin: other day for you3:29 Austin: You should probably make sure that that was something you did and not something else3:33 Austin: And that's because Git sends me emails3:36 Austin: And so it3:36 Luke: was3:37 Austin: still like like following my email account which I mean might be scary3:41 Austin: to some people but I didn't really care3:43 Austin: So like it's been sending me kind of these push notifications about emails that I have3:47 Austin: that it I should probably pay attention to which is kind of cool3:51 Luke: Yeah3:52 Luke: And I think there's like a of other modes in there too where you can like set3:55 Luke: goals even like you can go in there and say like I want to do this or I want to do that3:58 Luke: And it will like help you take actions on that goal specifically4:02 Luke: So4:02 Austin: it's4:03 Luke: I think it almost I almost feel like that's how it's4:08 Luke: And so it might be like thinking that keeping your inbox cleaned and organized as a goal4:13 Luke: of4:13 Luke: so it's like persisting on that task which is kind of interesting because like4:18 Luke: you if you use you know Claude or Chachapiti or Gemini like they have memory4:23 Luke: and they'll remember like what you have asked for it before4:27 Luke: And it will like you know do things for you4:31 Luke: But what's different about Muse is that it has those push notifications4:36 Luke: And so it'll it suggest actions like I noticed that X Y Z4:40 Luke: so I think maybe we should do you this4:43 Luke: And like that I think is like a different layer that I haven't seen4:46 Luke: it's not that complicated but it is removing the of taking an action with an4:52 Luke: like there's a value to it4:54 Austin: It definitely does seem more autonomous than the other4:57 Austin: It's a different thing for sure5:00 Austin: it feels more like feels more alive to5:03 Luke: be a little5:04 Luke: The other part of it too that's that's a interesting is it like asks you to5:08 Luke: give it a which I don't usually give my AIs a5:11 Luke: I just5:11 Austin: like create5:12 Luke: them5:12 Luke: But like giving your AI a feels like anthropomorphicization or I don't know5:18 Austin: always5:19 Luke: stumble over that word but you5:20 Austin: know5:20 Luke: what5:22 Austin: Anthropomorphicization5:23 Austin: Yeah5:23 Luke: Yes5:23 Luke: We'll call5:23 Austin: it that5:24 Luke: right5:24 Austin: Yeah5:25 Luke: Close enough5:27 Austin: Anthropomorphicifying5:28 Luke: It's close enough for our AI generated transcript to know what we're saying I5:32 Austin: Hopefully5:33 Austin: Yes5:34 Luke: exactly5:35 Luke: So5:35 Austin: yeah5:35 Luke: it's5:36 Austin: pretty cool5:36 Austin: I'm not sure what to have it do next5:38 Austin: Oh like one thing I did have it do5:40 Austin: The very first thing besides the email is I was at a5:44 Austin: do different like service jobs5:46 Austin: And so I already knew where I was going to go next based on I was like really up5:51 Austin: I live in Maryland and I was really up close to at this one job5:55 Austin: And so I wanted to do one more job in the day and I already figured out which one would6:00 Austin: be like kind of the6:01 Austin: And so I had Muse log into our service like service ticket management program and look6:09 Austin: at all my and I was like tell me which one I should go to next based on my location6:14 Austin: And it had all this reasoning of like well this one you have an scheduled6:18 Austin: for today but it's much further away6:21 Austin: And so basically it had all this like behind it6:24 Austin: And I was like oh this one is the closest but you don't have appointment until next6:28 Austin: week6:29 Austin: And so it ended up the one that I had already chose to go to6:32 Austin: And so it's cool6:33 Austin: Like I just wanted to see what it would do6:34 Austin: And it picked the same one that I did6:36 Luke: So that's pretty cool6:38 Austin: That's kind of cool6:39 Austin: And so it seems pretty helpful6:40 Austin: Okay6:40 Luke: as you're that story I remembered something else that I asked Muse to do that6:43 Luke: I6:43 Austin: was6:43 Luke: pretty shocked6:44 Luke: So I have a that I'm working on where it's like a website for I think I6:49 Luke: this before for the disc golf club that I'm6:52 Austin: a6:52 Austin: right6:53 Austin: Yeah6:53 Luke: Okay6:54 Luke: So one of the things that happens and the reason I want to make a website in the first6:57 Luke: place is that a lot of people in6:59 Austin: the7:00 Luke: club don't have Facebook but like is7:03 Luke: using Facebook to post like that are happening7:06 Luke: So what I did was I basically said because I'm like okay Muse is made by Meta7:11 Luke: So it's connected to7:12 Luke: It's got to know what to do here7:14 Luke: I was like how do we get the Facebook posts like onto the website7:18 Luke: Like what's7:19 Austin: the best7:19 Luke: way to do that7:20 Luke: So it basically created a script on its computer7:24 Luke: Like it has its own computer that I can use7:27 Luke: It had a and that script basically was listening for posts that come in7:33 Luke: And when a post is created that script pushes to right now the site is on it's on my GitHub7:40 Luke: and it's on GitHub pages7:43 Austin: it's7:44 Luke: not it's just like I'm just like working on it there so I can see it on my7:48 Luke: URL7:48 Luke: But anyway so it pushes a packet to my GitHub7:52 Austin: and7:53 Luke: then the website reads that7:55 Luke: and it is able to see it7:58 Luke: Eventually7:58 Austin: it kind of like an RSS feed sort of8:00 Austin: Basically8:01 Austin: to8:01 Austin: Yeah8:01 Luke: Yeah8:02 Luke: So it created that for me for the page that I'm trying to have it do that8:06 Luke: And it's it like I used a fine grain token so it has access8:12 Austin: to just that8:12 Luke: one repo8:13 Austin: Yeah8:13 Austin: I was wondering about that8:15 Austin: I was wondering8:16 Luke: And so it expires in 30 days and it has an set to like remind me like three days8:20 Luke: before it so I can like reapply for a new token if I8:24 Austin: want to8:24 Luke: keep it going8:25 Austin: Right8:25 Austin: I did that with the credit card report thing8:29 Austin: had a fine grain token so I can push updates through GitHub to like I don't have I only8:34 Austin: have one other person using it at the moment but I plan to have like as many people at8:39 Austin: my company as I can8:40 Austin: And so yeah I was a bit concerned about the fine grain token expiring because8:45 Austin: I said it's like a8:47 Luke: Yeah8:47 Luke: No that makes8:48 Austin: sense8:48 Luke: I could I could extend it but I again I don't necessarily want to extend more like8:54 Luke: I think 30 days is adequate8:56 Luke: It's doing what it needs to do and I can just give it a8:58 Austin: new one8:58 Austin: Yeah8:59 Austin: So8:59 Austin: Well I'm not too concerned because it is fine grain9:01 Austin: So it only has access to9:03 Austin: Yes9:03 Austin: It only has access to not only the repo but I like the releases9:09 Luke: Right9:10 Luke: Right9:10 Austin: So it's not like it's not really that dangerous9:13 Luke: Yeah9:13 Luke: Yeah9:14 Luke: Well and if my9:15 Austin: disc9:15 Luke: golf site if my disc golf site crashes it's fine because it's9:18 Luke: not live right now9:19 Luke: It's still a9:19 Austin: Yeah9:20 Luke: If it goes into production I got to figure it out more9:22 Austin: Right9:22 Austin: So one thing that I've been feeling is like I'll be driving around or doing something9:27 Austin: and like I feel this like itch that I should have something running in the background9:32 Austin: Like I should feel like I'm wasting time if I'm not having an AI do something for me9:37 Austin: which feels a bit like a drug addiction maybe9:41 Austin: But it's9:41 Luke: It is9:42 Austin: It's just feeling like I'm like man I got this9:44 Austin: I should be doing something right now9:45 Austin: Or if my usage runs out that definitely feels like a drug addiction because I'm like oh9:51 Austin: man I just need to do something but I have tokens9:55 Luke: Yeah9:55 Luke: So I've been thinking about this too a bit and I feel like there is a really important10:00 Luke: thing to think about here because the models are going to get better and and better10:06 Luke: And there is going to come a where they start to do things like our let me try to10:12 Luke: put it way10:13 Luke: I think our moat like humans in the way that they do work and the way that they interact10:18 Luke: our moat is like our humanity which sounds like really cheesy but it's true10:21 Luke: So it's kind of like I know if you're like into gaming culture or but10:24 Luke: you hear people say like touch grass10:26 Luke: Like I feel like the most important thing about AI is to know how to use it incredibly10:31 Luke: well but then also touch a of because10:35 Austin: you have to stay in reality10:37 Luke: A percent10:38 Luke: You have to be human and understand what's going on because eventually the AI is10:42 Luke: to be so good that it's going to be taking10:44 Luke: And I feel like with this next level of like AI that we're experiencing like with Astra10:49 Luke: with Fable 5 1 we're hitting this point where I'm starting to like realize that the stuff10:56 Luke: that's in my mind that it's like a vision for what I think I can see like my systems11:00 Luke: and things that I'm to make it's going in that direction really quickly and faster11:06 Luke: than it was before11:06 Luke: And so I'm like okay I can see the fact that this is getting better and better like11:11 Luke: what's next11:11 Luke: And I feel like that's a scary11:14 Luke: But the other part of it is as the AI gets better my skill in using the AI becomes less11:20 Luke: relevant11:20 Luke: And what becomes more relevant over time is my ability to stay grounded in my humanity11:26 Austin: Yeah11:26 Austin: Yes11:27 Austin: So11:27 Luke: like I think it's a transitionary period where like there are some people that are11:30 Luke: better with AI than others11:32 Luke: Eventually the AI is to be so good11:33 Luke: You don't have to be good to use it11:35 Luke: Yeah11:35 Luke: It's11:36 Austin: definitely the barrier11:38 Luke: to11:38 Austin: entry is getting lower and11:40 Austin: I think they're gonna have to a new word like past exponential11:44 Austin: Like is there something greater than exponential11:47 Austin: Well I11:47 Luke: think exponential feels like it's not enough to describe it but I think that11:53 Luke: is kind of the to a extent because it's like the growth rate itself is11:59 Luke: exponential not just like the you know what I mean12:01 Austin: Everything about it12:02 Luke: Yeah12:02 Luke: So at it feels that way12:04 Austin: Yeah it's pretty crazy12:06 Luke: Because we12:06 Austin: used to12:07 Luke: like model updates like every three months and now they feel12:10 Luke: like they're every month12:11 Luke: And12:12 Austin: each12:12 Luke: model update is like so significant12:14 Austin: Yes12:15 Austin: And speaking of which there's a new model or new concept for AI12:20 Luke: Yeah like a new whole new type of AI12:22 Luke: Like I don't even think12:22 Austin: it's12:23 Luke: MLM is it12:24 Luke: No it's12:25 Austin: not12:26 Austin: LLMs are large language models12:28 Austin: that's what everything else AI is12:31 Luke: basically12:31 Austin: take12:32 Luke: Yeah you hear people say AI you're hearing them say large language model almost 9912:38 Luke: of time up until this point where there's now actually a new type of AI12:43 Austin: Yeah so apparently LLMs are system two models12:46 Austin: And this is a system one model12:48 Austin: I honestly don't know what that means12:50 Luke: Okay I do12:51 Luke: So I'll explain it really quick12:53 Luke: So there's two types of thinking12:55 Luke: There's system one thinking and system two thinking12:58 Luke: two thinking is like when you have to do long division you take out a13:02 Luke: of and paper and you write it you do the carefully13:06 Luke: System one thinking is when I say what's two plus two13:09 Luke: You really think four13:10 Luke: There's no thinking at all13:12 Luke: It just automatically comes because it's part of your actual reflex13:16 Luke: So it's almost like a reflexive thinking is system one and then actual spending caloric13:22 Luke: thought is system two13:24 Austin: Right13:25 Austin: So this company called TypeSafe has just come out with a system one model called Jev13:33 Austin: And the big thing about it is one it's insanely cheap because you're only paying for13:39 Austin: the input tokens13:41 Austin: And the input tokens are like 0 072 cents or something13:46 Austin: Or even not even I don't even think it's one cent13:48 Austin: It's like a fraction of a13:50 Austin: And so you have your input13:52 Luke: which is13:52 Luke: like a fraction of per million tokens13:55 Austin: I think so13:56 Luke: Usually per million tokens is what they go by13:59 Austin: I'm not sure14:00 Austin: Something like that and14:01 Luke: it's insanely14:01 Austin: cheap14:02 Luke: But it's extremely cheap14:03 Luke: Yeah for sure14:04 Luke: Yeah14:04 Austin: it's crazy14:05 Austin: So basically instead of an LLM you're it all this input it all these tokens14:10 Austin: thinking and then it spends tokens outputting you a of text14:14 Austin: What it's doing is you're basically giving it a multiple choice question and it gives14:19 Austin: probabilities it probabilities of which answer is the correct one14:23 Austin: So there's a of different implications on what you can use this for14:26 Austin: And at the basic level of like every piece of is an if then statement14:32 Austin: So it's basically just a bunch a series of a of questions that are yes or no14:36 Austin: like or they have multiple answers14:40 Austin: And so Jev can process these type questions extremely fast and extremely cheaply14:46 Austin: And like so I for the first thing I did with it was I've always been interested14:51 Austin: in14:52 Austin: so I set up I had Cursor or no I think I had Hastra set up a trading platform that14:59 Austin: uses Jev15:00 Austin: And so that's basically it decides the probability of if you should buy or wait15:06 Austin: And all I'm it is I have all the candles that are from Coinbase and15:11 Austin: my script calculates different indicators15:14 Austin: so it that information to and Jev gives the probability of which of the15:20 Austin: three decisions is the best one to do15:23 Austin: And I've had it every minute and I've had it for I haven't checked15:27 Austin: in a but the first eight hours I had it it hadn't even spent a dollar15:31 Austin: It was like I don't even know if it had spent 10 cents15:33 Austin: And it a of and it used like a tokens already15:37 Luke: I just checked I just checked the price while you were giving that explanation and15:41 Luke: it's actually 4 2 cents per million tokens15:45 Luke: That's15:45 Austin: crazy15:45 Luke: Like you could spend like a ton of and not even get15:49 Austin: a15:49 Austin: And so you have to be on a wait list for it right now which Luke just got on the wait15:53 Austin: list15:53 Austin: I had just we're both kind a little bit late to the game because it came out it just15:58 Austin: came out last Tuesday16:00 Austin: But which is only like less than a ago but everyone's all talking about16:04 Luke: it16:04 Luke: we hopped16:06 Luke: Yeah16:06 Luke: In the AI world a week ago is late to the game16:09 Luke: I16:10 Austin: know I16:12 Luke: So16:12 Austin: yeah so I got on Friday night I think I put in got on the wait list on like Thursday16:17 Austin: night and I got in on Friday night16:20 Austin: And so Saturday morning I was building this trading thing and it's pretty crazy and it16:25 Austin: feels different16:26 Austin: And everyone's talking about how crazy it is16:29 Austin: And the the that made it helped co found ChatGBT and various other companies16:34 Austin: And they said they'd been like developing this for the past two years and just like16:40 Austin: suddenly dropped it on Twitter on16:42 Luke: Yeah16:42 Austin: And everyone's freaking out about it16:44 Luke: It's kind of crazy because I mean the about it is like it's it's an LLM as like16:51 Luke: a software component16:53 Austin: almost16:54 Luke: Like right now we think of the large language but it's not an LLM16:58 Luke: It's an AI16:59 Luke: That's like a component17:00 Luke: like right now when you interact with LLMs like they're often doing stuff and like17:05 Luke: I've even done like classification and like read this thing and give me like what category17:12 Luke: it to and what probability that you're like stuff that Jeff should be doing17:16 Luke: Right17:17 Luke: I've done this type work with an LLM and it's just a little bit messy17:22 Luke: It's slow17:23 Austin: It's not17:23 Luke: fast17:23 Luke: A17:24 Austin: lot17:24 Luke: of tokens17:24 Austin: are17:25 Luke: wasted17:25 Luke: A lot of tokens are wasted17:26 Luke: You're a lot of for those tokens17:28 Luke: And it's not actually that computationally significant to do what it's doing but the17:33 Luke: models are trained to do text and to like talk back and forth and have pros and you17:38 Luke: know reinforcement learning with human feedback17:41 Luke: And what does the human want to see17:43 Luke: That all gets thrown completely out the and it literally just processes context and17:48 Luke: gives you an answer17:49 Luke: That's it17:50 Luke: it's like it's been the probability17:53 Luke: So what does that mean17:54 Luke: That means like I think where this goes is like an agent is going to call17:59 Luke: Jeff as a18:00 Austin: Yeah18:00 Austin: So that's important to note18:02 Austin: It's does not replace LLMs by any means18:05 Austin: And it would be used like in the sense of I've heard of the example of like a customer18:08 Austin: service situation where a frustrated customer sends a to a and the company18:16 Austin: has Jeff instantly say like how mad is this person18:21 Austin: And like it might you might say like you might give it give it the message and then18:25 Austin: give it like here are the options like not mad extremely mad very mad18:31 Austin: angry or whatever18:32 Austin: And it will give the probability of each18:34 Austin: And so if it decides that it's super angry then it could hand it off to like an actual18:38 Austin: person18:39 Austin: But if it that they're not really that then it can hand it off to an18:42 Austin: LLM to respond to it18:44 Luke: I could even decide like what is the probability of a satisfactory resolution if18:49 Luke: I hand off to a18:50 Austin: Right right18:51 Austin: Yeah yeah18:52 Austin: So18:52 Luke: could even decide if the handoff is worth it or not18:55 Luke: And18:55 Austin: since it's so cheap18:57 Austin: It can be used to make so many deterministic18:59 Austin: Answers that is like the foundation of19:03 Austin: And so you can have it you can have Jeff pretty much checking every step and routing to19:08 Austin: different things19:09 Luke: So in every everything like everything19:13 Luke: And that's so it's called Jeff19:15 Luke: And that's an interesting term19:17 Luke: So Jeff is short for Jevons paradox19:22 Luke: I19:22 Austin: didn't19:23 Luke: know that is19:23 Luke: Yeah so so their idea behind this is that because it is so ubiquitous and it's like19:30 Luke: so useful in every situation and the price of it is so cheap that everyone's going to19:36 Luke: use it everywhere is19:38 Austin: right19:38 Austin: Yeah19:38 Luke: So like you're just going to pay you're just going to pay like an extra 2030 bucks19:42 Luke: a because you're on everything19:44 Luke: Right19:45 Luke: There is Hank Green has19:47 Austin: a on the paradox which is pretty good19:50 Austin: So if you want to know more about19:51 Luke: that go19:52 Austin: look up19:52 Luke: Hank19:53 Austin: Green19:53 Luke: Jevons paradox and just19:54 Luke: watch this video19:55 Luke: It's pretty good19:56 Austin: All right that's going to do it for this week19:58 Austin: Thank19:58 Luke: you guys for and we will see you guys20:00 Austin: next week
AI Meta Muse TypeSafe Jev Technology
Highlights
[0:27] Luke introduces Muse, Meta's new AI agent.
[1:12] Discussion of Muse's permission-based task handling.
[2:38] Austin shares his experience using Muse for email management.
[4:31] The unique push notification feature of Muse.
[9:23] Discussion on AI's accelerating development and its implications.
[12:23] Introduction to Jev, TypeSafe's new AI model.
[13:53] Jev's cost-effective processing and potential applications.
[19:22] Jevons Paradox and its relevance to AI usage.
Links
Follow Us
