I Tested Jev on 12 Real Use Cases. My Honest Thoughts.

0:00 So, Jev is literally everywhere and I think it's going to change how AI automations are built. So, I came in here and tested it on 12 use cases and compared it with other AI models on

0:08 things like speed and cost. I was even able to build this Chrome extension that will label X tweets as breaking or, you know, golden nuggets or AI slop in real time for me. On the back end, you can

0:17 see that it uses Jev to actually instantly categorize all this stuff as soon as it enters my screen. It's not doing very well in the first hour, but I also made this Jev trader which is

0:26 literally every single second analyzing if Bitcoin's going to go up or go down or stay. And then it basically places trades in real time for me because this model is so good at quick decisions. But

0:35 anyways, by the end of this video, you'll understand how Jev works and where you should actually use it in your life. So, let's not waste any time and just get straight into this one. All

0:41 right, we're going to start off with just like what is Jev. I'm not going to do a super super deep dive, just enough for you to understand how it works and what we're looking at in today's video.

0:48 So the interesting thing about Jev is that it's an AI that makes decisions, but it doesn't write anything. It doesn't output tokens. It's not anything that you could actually have a

0:56 conversation with. It just makes decisions. So here was kind of the announcement tweet from Dogo. He co-invented CHABT and then he has been building in the past 2 years this new

1:06 way to train models, RLCD. And you can see what that stands for is reinforcement learning for calibrated decisions. And this is on Typesafe's blog. And by the way, if you want to

1:16 actually get in here so that you can start playing around with Jev, then go to Typesafe AAI and join the wait list and then hopefully in a few hours you're able to sign in. But also, this is

1:24 available through like Versel's gateway as well as open router. So if you're not in the wait list yet, then you can still go out and play with Jeff. But anyways, essentially what happens is instead of a

1:33 normal chat model where you would send in a message like this, this is some sort of support ticket and the AI model would read it would reason, would think, and then output like a message or output

1:42 some sort of classification. It basically just outputs these types of things which are a yes or no confidence level, a category, and sort of a score. So for the first one, is it urgent? 99%

1:53 confidence is yes, it is urgent. Which team? There were probably multiple routes like technical or billing or support and it labeled it as technical. and then how frustrated. It gave it a

2:01 one out of two on the frustration score or scale. But you're fully in control. You basically will set up Jev with, hey, this is essentially how you're supposed to make decisions and here is sort of

2:10 like the classification criteria. So, it's three types of decisions. Like I said, the yes or no is called a new. The pick one is a choice. And then we have an actual score. And I'm going to show

2:20 you guys real examples of all of these being run on these 12 use cases. So, don't worry. But I just wanted to sort of lay the foundation here. So, like I said, in this example, it's yes or no.

2:28 And there's a confidence score in the team or categorization example. It's different categories as well as a confidence score and then a score from 0 to 10 on some sort of scale. And in

2:38 here, one meant that they were frustrated. And the reason why this is getting so much traction is because Dio said that this is 20 to 200 times faster and 40 to 400 times cheaper with output

2:48 tokens being free. And so if we look at the speed here, compared to models like Terra and Luna and Soul, this thing is going to be a lot faster. This was just one very quick test I ran. This doesn't

2:58 mean that Terra is always faster than Luna, but this was just one quick example of how significantly faster Jev is. Once again, it doesn't have to output all these tokens or reason. It

3:07 just boom makes a decision and outputs it in like this JSON format. And same thing from a cost perspective. If you were running thousands and thousands of decisions per day, this is what it could

3:16 actually end up looking like. Now, obviously, the important thing is you're paying a lot less. So, you want to make sure that the quality is the exact same as what you'd be getting here or here to

3:23 justify it. If we're just looking at the cost right now, it is significantly cheaper and it is proven that it's significantly cheaper. So, what it cannot do is write or summarize or find

3:32 themes or do deep analysis. It basically just outputs decisions. And one other limitation right now is that it has a very small input context window. It's 64,000 tokens, whereas a lot of the

3:43 models that we're used to using today, whether that be Claude or GBT, are more on the side of a million tokens. So, if we look at this on a use case like YouTube comments, if I fed in 5,000

3:52 YouTube comments, Jev could sort them for way cheaper and way faster than any AI model could. And then it could categorize them as like, "Hey, these need a reply. These are stuck. These

4:01 people want to buy." And then what you could do is feed a more intelligent AI model, an AI model that actually responds to things and outputs things. And you could say, "Hey, Chadbt, could

4:09 you read just these now and then help me like analyze themes or help me respond to these ones or something like that?" And that way you're not using a slower and more expensive model to actually

4:18 categorize all of those comments in the first place because Jev isn't a frontier AI model. It's not even in the same bucket as Astra or Fable. It's a completely different type of model. So

4:27 when to use it if you have thousands of things, if you have a corpus of data, if you need to classify things, make decisions or you're running some sort of decision- based or classification based

4:34 workflow at scale in production. if you need like very quick realtime decisions because it's really really fast and stay with chatbt if you need things like a handful of items you need to understand

4:44 why you need to brainstorm you need to chat things like that anyways let's just get straight into some use cases here so I know this might look a little bit intimidating of a screen but what I want

4:52 to show you is a little bit of a playground of how this actually works by the way guys I've got this completely free SOP for you about getting your first aation client it's going to go

5:01 over the exact steps that has been proven for hundreds of our AIS plus members to get their first paid gigs It goes over the one sentence service pitch that can get you started today. Why your

5:09 first client should cost you money. The five-minute video that answers can this person actually deliver before you've actually received any money. What to do when you have zero case studies. There's

5:17 so many good things in here that are going to help you out. Even if you already do have clients, I would recommend grabbing this because like I said, it's yours completely free. So, if

5:25 you want to grab this, there's a link for it down in the description. Let's get back to the video. So, the first one we're looking at is emails. Now, real quick, you can see that I've got a bunch

5:32 of different categories set up or a bunch of different questions set up. The first one is invoice or receipt. This is a yes or no. The second one is brand deal. This is yes or no. Scam or

5:40 fishing, yes or no. We also have email type. We also have urgency. And we also have sponsor fit. So, we've got different types of scoring and different types of categorization. And you can see

5:48 here if I run this real quick, if I go to redo everything, and I hit run, we're currently using the model Jev. And this is a,000 emails. Like look how quick this is able to classify a,000 emails.

5:58 So, it did that in about 70 seconds for 9. Now, obviously that's not like super super fast, like lightning fast, but this was not parallelized. If we were to run all of those individually in

6:08 parallel, it would have been so much faster, but I just wanted to show the difference here. Let's even go to something like GBD 5.6 Luna. And we'll run this again on everything. So, all

6:16 a,000. I mean, this feels like it took forever. With Luna, that took 5 minutes and it was 62 cents compared to 70 seconds and I think it was 9 cents. And also, think about it wasn't just doing

6:27 one sort of classification. It was doing all seven of these rules. And I actually just changed the back end to make this actually process things more in parallel with bigger payloads. So I'm just going

6:35 to run this now and you'll see how much quicker this really is. Boom. Look how fast that went. That took 6 seconds and it once again was 9 cents. So that just shows how you can optimize that backend

6:44 to make Jev go even faster. And yes, you can do the same thing with Luna, but it's just not going to be as fast as 6 seconds for a,000 emails across seven categories. So I know that this

6:52 interface may be a little overwhelming. Let's just get rid of everything here except for invoice or receipt. So, this is literally just us saying, "Okay, we want to set up some sort of

6:59 classification for all of these thousand emails." We're going to give it a name. We're going to ask the question, is this email a receipt, invoice, payment confirmation, or billing notice? And

7:07 then we just define like what counts as yes. So, it has a a charge, a payment, a payout, or some sort of failed payout. And we call it a yes when Jev is at least 50% confident. So, that's kind of

7:17 the thing that we're running here. And I would just go ahead and choose redo everything. And so, when we run this, it's basically going to look at all a thousand of those emails and just

7:23 decide, is that an invoice or receipt? It comes back in 4 seconds for five cents. And where it landed was no on 763 of them, but yes on 237 of them. Here is a category example where we actually set

7:34 up the email type, whether it's notification, newsletter, billing, opportunity. And you can see if I edit this, what we did is we actually had to choose the options and define each of

7:42 them. So this is a pretty standard AI classification type of automation. But then if we look at the scoring, so for example, if we look at the um sponsor fit, if I go to what this one looks

7:50 like, this is rating it on a scale. And we basically choose from lowest to highest what that looks like as far as not a sponsorship inquiry at all or if it's a strong fit. And it will choose

8:00 the level. So you can see here it shows that 942 of them were one and then it, you know, we didn't have any that were strong fits. You can see there was another score one that was urgency about

8:09 if there was action needed or nothing needed at all. And this was a little bit more even across the board. The average was 2.8 out of five. So anyways, email classification is one example. And

8:19 obviously when you're looking at the actual cost and speed here, that first example we ran, Luna was 12 times the cost and 46 times the time and that was also just Luna. What if we went up to

8:29 Terra and Soul? How much more expensive that would have been and how much more that would have cost us. So a lot of these use cases are kind of on the classification side. I did the exact

8:36 same thing here with YouTube comments where I can choose categories or yes and no and score. And I can analyze thousands of comments at a time on things like comment type, if they're

8:45 worth a reply, if it gives me a video idea, what the sentiment is, what the question difficulty is. And once again, if I run Jev on all a thousand of these, look how quick that actually goes. It

8:54 would be so much longer if we used any other sort of AI model. 5 seconds for 5 cents. And if we actually go to my Jev console real quick and I go ahead and refresh, this is going to show that I've

9:03 used 85 cents with Jev. But look how many requests I've done. I've done almost 20,000 requests. Think about what 20,000 requests on a different AI model would have cost us. And I can also do

9:12 the exact same thing with my school posts. I can set up my custom categories or my custom classification criteria to see what type of questions we have. If they need help, if we need a team

9:21 answer, if there's churn risk, member experience level, testimonial strength, and yes, this is kind of like a dashboard playground view, but what if you had this in an actual automation

9:30 where every single time a new post got made, you updated the database. Every single time a new YouTube comment came in, you updated the database. every time there's a new internal CRM entry or

9:38 every time there's a new lead that submitted a form on your website. There's so many things you can do here. And even though the speed might not not matter a ton when it comes to like

9:46 actually having automations in production, what does add up really quickly is the cost because once again when you start to run thousands and thousands of requests through, it's

9:54 going to add up big time. So when you talk about AI economics and model routing, this is definitely going to be a game changer. Now here's another interesting one where the speed really

10:01 did matter. I have this one called X feed where it basically like pulls in a bunch of posts on my feed and it will tell me are they on topic or you know like what's the category? Is it breaking

10:10 news? Is there video idea potential? So similar classification as we saw in these first three. But look what else I did. If I give my X a hard refresh real quick, you'll see that in the bottom

10:19 left I have this thing called Jev judged. And you can see that it judged that one a slop. And as I scroll through, this is a Chrome extension that it built for me where it's looking at

10:27 the X posts and really quickly reading them and classifying them as breaking or golden nuggets or AI slop. It's basically going to help me keep me more focused while I'm scrolling through X to

10:38 see, you know, like highlighting what might be good to read and what are things that I should probably just ignore because of AI slop. And this was just a Chrome extension that I built

10:46 where it has Jev on the back end powering all of this. So, that is a pretty cool use case and it shows how fast this thing actually happens in production. I mean, look how fast it's

10:54 reacting to these posts as they come onto my screen. It's basically instant. I also thought about what you could do with your meetings here. You could analyze meetings as they get transcribed

11:01 in Fireflies or Granola or whatever you use. And as soon as they come in, you can categorize them by what type of call it was, if you had decisions made, if you had like action steps. I thought

11:09 this one was an interesting one because let's say you see that in a lot of your calls you have no next steps discussed or defined clearly with ownership and timelines. then you can use the data to

11:18 change how you're conducting your calls. So, what I think is interesting is Jev on its own doesn't analyze things for you, but if you're strategic with the way that you set up the questions, you

11:27 can get analysis from it. You can have this data tell a story. You can't have Jev look at thousands of transcripts and say, "Hey, tell me what I need to do better about these meetings or tell me

11:34 common themes." But what you can do is you can give it categories and you can give it scores and then from all of these different types of questions that you set up, you tell your own story with

11:43 that data. You can see that there's not much tension in some of our calls. I mean, this one has someone says they're frustrated or overloaded, but but all of these questions that I created for Jev

11:52 here, there's some sort of takeaway from each of these. Are things waiting on Nate? Is there revenue relevance? Things like that. And I also tried this use case with video clips where I basically

12:00 had Jev or I had, you know, Astra break up a bunch of my YouTube videos into clips. And then I had Jev look at those clips and tell me, you know, is this possible to be posted on its own as a

12:09 clip? A lot of them, no. What is the hook strength of these clips? What type of clips are these? Do I need the screen on them? Is there a quotable line inside of this clip? And then I could start to

12:17 have it pull out things that might be worth reposting somewhere else or might be worth me thinking about the way that I actually speak in these videos and things like that. And so these were a

12:26 lot of ways that I would think about how do I take a corpus of information, a ton and ton of data that I want to have AI analyze, but instead of paying more for it and waiting longer, let's figure out

12:36 how we can use the right questions to have Jev do it for us. And then not only can we maybe have some sort of dashboard view, but how do we build Jev into our actual back-end automations where it's

12:45 really going to benefit us to having a really cheap and fast model as we increase the throughput. Now, I will say don't just plug in Jev and trust what it says automatically. What you're really

12:55 going to want to do is run evals, meaning you're going to have a golden data set of 100 use cases and 100 correct answers. And then you're going to run Jev through those and you're

13:03 going to run Opus through those. and you're going to run soul through those and you're going to see which model gives you the best balance of accuracy and cost. And if you care about speed in

13:11 that use case, then also speed as well. But here are some other things you could do. You could have it vet contracts for you as they come in. You could see the type of risk. You could see the type of

13:18 clause. You could create any sort of questions that actually matter to you when you're vetting contracts. Same thing with jobs and leads. You can get red flags. You can get lead quality. You

13:27 can get next steps. You can also do something like a brain dump router where you're constantly just talking into your phone or you're talking into something and then you're feeding that into Jev

13:35 and it can tell you what type of things you're talking about. If they're ideas or tasks or journals, if you have things that have deadlines, if you have things that are high priority, what area of

13:44 your life they're in. There's so many ways to use this because a big part of what we do with AI, like I said, is just figuring out what to do with all of our data. And Jev can do that really well.

13:52 And then I think one of the best examples here is customer support. just routing emails around, figuring out, you know, sentiment, urgency, what we need to do, how we categorize these sorts of

14:02 things. I think that this is going to be a huge game changer for customer support because there's so many different like decisions that have to be made and Jev is really good and really fast and

14:11 really cheap at making decisions. So, here's another use case that I thought would be cool to just sort of like PC. So, this is paper trading. This isn't very vetted. There's a lot that's like

14:20 wrong with this, but I think that Jev being so real time and making decisions so fast, it's going to be really interesting to see how it affects things like day trading or trading crypto in

14:29 real time. So, you can see every single second Jev is basically predicting is this going to go up? Am I unclear? Is it going to go down? You can see these confidence scores jumping around every

14:38 single second. And that's how it decides what to do. That's how it decides down here to place trades, to buy things, or to sell things. Now, unfortunately, there's a lot of these fees here. So,

14:46 the fees are way more expensive than Jev actually um making decisions for us. So, that's one issue where it's like, okay, well, how much would we actually have to be able to profit to make this worth it?

14:55 But right here, you can see just the cost of running these decisions. Look how much more this would have cost us per day with other models, whereas Jeff would just be costing us about two bucks

15:01 a day to run this 24/7. And Soul and Opus and Fable would be significantly more than Jeff. Now, I've also seen people on X doing things like having Jeff play video games and having Jev

15:12 like do different creative fun things. And I think it's really cool like the browser use and all that. It's cool to see what's possible. But I think you have to think about where's the handoff

15:19 because with browser use it was unable to like actually type things in. It would basically make decisions and it would have to route to a different model that's better with actually controlling

15:26 the browser to do things. And that's why I wanted to show what this looks like for these examples because I think building dashboards or automations where you have Jev powering it on the back end

15:34 for things that you actually care about in your life like these sorts of things is where you'll start to play with Jev and actually get some return and then later figure out how you can expand or

15:43 extend your workflows with other models on the back. But anyways, I hope seeing these examples, even though all of these were very similar in the, you know, the realm of classification, I hope that it

15:52 helps you understand how you can start to ask the right questions in here and how you can start to work it into things that you're doing to actually make sense out of it. But that is going to do it

15:59 for this one. So, if you guys enjoyed or you learned something new, please give it a like. It helps me out a ton. And as always, I appreciate you guys made it to the end of the video and I'll see you on

16:06 the next one. Thanks everyone.