Stop Using ChatGPT for Everything—Jev Is Here

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Stop Using ChatGPT for Everything—Jev Is Here · Channel: Eric Tech · 08:14 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Markets and economy, AI models

Summary

The video introduces Jeff (also referred to as Jev in the transcript), a decision-making model that scores available actions instead of generating long text responses. The presenter says this can make it faster and cheaper for tasks such as classification, routing, games, browser control, and short-term trading, while noting that it cannot write, explain its reasoning, or perform step-by-step reasoning like a text-generating model. The video highlights public projects using the model, a browser-use demonstration, and ways to try it through Vercel AI Gateway or agent skills, though access may require a wait list. A sponsored segment demonstrates Co-Invest, a trading assistant that analyzes a portfolio and prepares trades for user confirmation.

Key points

Jeff is described as a decision-making model: it scores supplied options and returns a selected decision rather than generating text token by token.
The presenter says the model can make 10 decisions per second and demonstrates it playing Doom.
The video claims Jeff is cheaper than several named models and reports a low hallucination rate, while noting its intelligence score is similar to some mid-tier models.
Suggested uses include classification, scoring, routing, games and simulations, and short-term trading.
Jeff cannot write sentences or code, explain itself, or reason step by step, so the presenter advises against replacing all other models with it.
The presenter says a public repository lists over 40 projects built with Jeff, including classification and routing examples.
A browser-use demo shows Jeff helping navigate flight booking in under 10 seconds; access to the tool is described as wait-list based.
The video says Jeff is available to try through Vercel AI Gateway and can be called using its AI SDK or a curl request.
The presenter also mentions Jeff-related agent skills intended to help AI agents use the model for smaller decisions in applications and workflows.
The sponsored Co-Invest segment shows portfolio analysis and trade preparation inside ChatGPT, with trades requiring confirmation.

Chapters

Start	Chapter	Summary
00:00	Introducing Jeff, a decision-making model	The presenter contrasts Jeff’s option-scoring approach with text-generating language models and discusses its speed, cost, and reported performance.
02:05	Use cases, limitations, and project examples	The video outlines suitable tasks and limitations, describes community projects, and includes a sponsored demonstration of the Co-Invest trading assistant.
06:11	Browser control and flight booking	The presenter describes browser-use integration and a demo in which Jeff helps navigate a flight-booking site.
07:13	Trying Jeff through Vercel and agent skills	The video explains how to access Jeff through Vercel AI Gateway and mentions skills for using it in AI-agent workflows.

Quotes

Jeff here is going to basically make a decisions rather than generating a text response. — 00:31

Jeff here is able to make 10 decisions per second. — 01:02

It cannot explain itself. It cannot write a code. — 02:35

Jeff here is able to use browser use and be able to book a flight — 06:42

Concepts and entities

Jev, TypeSafe AI, Vercel AI Gateway, Co-Invest, Browser Use, Repositório de projetos com Jev

Full transcript (original language)

00:00 On September 15, there was a new model here released. It’s called Jeff. >> My name is Jeff. >> And we can see that ever since the model he released, you can see we got over a,000 points on hacker news alone. And just a couple days after, versel has launched Jev on their AI gateway. And this model here is created by this company called Typesafe AI. And the person who created this company is also co-invented chatbt. Now, this model here is really different compared to any other models that we’ve seen because this model here is not a text generation models. It’s a decision-making model. So in a normal LM call, you give it a

00:31 question and LM here is going to generate the response token by token and therefore the response here is going to be a lot slower. Here you can see that it took 8 seconds. And if you were to give the same question here to Jeff, you can see that Jeff here is going to basically make a decisions rather than generating a text response. And you can see that we’re simply providing the options that Jeff going to make a decision. And Jeff here is going to score each decision here with a score. And whoever has the highest score is going to return that decision back to the user. So you can see the response time here is a lot faster because Jeff

01:02 here doesn’t have to give a really long text response. And here you can see that we have a quick demo where Jeff here is playing the game of Doom. And you can see that Jeff here is able to make 10 decisions per second. So fast that this model here is able to play a game like this. And according to their official websites, you can see that Jeff here is actually cost way less compared to models like Haiku or Opus 5, Sona 5 or Luna. And if you were to actually look at the intelligence score for this model, you can see that it’s getting a very similar score compared to Terra or Sauna 5. And if you look at the

01:34 hallucination, you can see that Jeff here pretty much have 0%. And what’s really interesting here is you can see that for the input and output tokens, Jeff here is completely free for the output tokens because it doesn’t really generate anything. and just make a decisions and that’s it. And here if you were to look at Jeff for the 1 million tokens per inputs you can see that that’s how much it costs 0.042 042, right? And if we look at OpenAI like Astra, Fugo 5, you can see that’s about $5, $10, you can see that’s extremely expensive compared to type safe Jeff. Okay, so now you know what’s good about

02:05 Jeff. Let’s take a look at what Jeff is really for. Now, like I said before, a clear example could be games or stock trading, day trading, right? Something that you need to make a decision fast and be able to get a decisions right away, right? So you can see the Jeff here is really good for this. Now there’s comes to tradeoff. So Jeff, you can see it can be able to pick an action, it can be able to classify things, score things, and can be able to get results faster than any other large language models that we tested before. But what it cannot do here is that it cannot write a sentence. It cannot

02:35 explain itself. It cannot write a code. It cannot be able to reason things step by step like any of those models that we tested before. So before you think about, okay, well, let’s throw all the models away and just use Jeff. Well, think again. Because Jev has a similar intelligence score like models like Terra or Sonet 5. They’re not the deep frontier models like Soul or Opus 5, Fable 5 or, you know, like Astro 6 or whatever, right? Those frontier model here requires time to do deep thinking. But this model here is meant to do

03:06 something quick and it only outputs decisions rather than just a long text outputs. Now, if you actually want to try out Jev, you have to be on the wait list. And currently, you can see I’m on the wait list right now. And of course, if you’re interested for me to test out Jeff, make sure to comment down below. What are some ideas that you want me to test out with Jeff, right? What are some use case that you want me to test it out? And I’ll make sure to make a video on that in the near future. And of course, don’t forget to subscribe for this channel here because you will be the first one to notify once I have a video out on a real test cases for Jeff. But now, before you click away this

03:36 video, there’s actually a repository which has a curative list of public projects that people are currently building with Jeff. And currently you can see there’s over 40 projects that people have building stuff with Jeff and large portion of this from classification and routing. So let’s say if you’re currently doing code reviews and you want to know exactly which model to use or which tool to use and for this kind of quick decisions, we can literally automate this using Jeff or even be able to route to whichever skill you want to use, right? Those are all can be automated using Jeff to help you to save more tokens whenever you’re

04:07 using different large models. And here you can see we can also do this for game and simulations. So, let’s say if you want to do drones or playing games or Pokemons, right? You can also automate this using Jeff. And a really nice one here, you can see, like I said before, financial and short-term trading. You can now be able to automate short-term day trading here with Jeff, which is really, really cool. Now, speaking of trading, let’s bring up the sponsor of this video. I want to show you something I’ve been testing recently because it solves a pretty annoying problem when you’re actually managing a portfolio. You might already have a bunch of

04:38 positions, then some major news comes out. Now you have to figure out which parts of your portfolio are actually affected. That’s basically what co-invest is useful for. It’s a trading assistant that connects directly to your account and lets you analyze your portfolio and prepare trades from inside chatbt. Plus, setting it up is pretty straightforward. I just create my liquid account, connect co-invest through the chatbt app directory, and once it’s connected, I can start working with my portfolio directly from the conversation. So, I’ll just ask it to analyze my portfolio based on today’s market news. And you can see it pulls in

05:09 the account context and starts going through the positions. It can look at things like current market positioning, funding rates, and recent catalysts, then connect those back to what I’m already holding. It tells me which positions are the most relevant. Now I can ask it to actually suggest some adjustments. And instead of me manually going through every headline and position, it turns that analysis into specific orders that I can review. I can see the asset direction, size, and reasoning right here in the chat. The connection can read the account and place trades. as you confirm, but it can never withdraw or move your money. After

05:41 it’s all set, then we can place a trade based on the change in the portfolio, which is opening a trade for smaller tactical positions. I’m selecting cost here, and then I’ll hit confirm. You can immediately check the change by asking Chat GPT to view my portfolio as well. So, if you want a quicker way to build a portfolio and manage it in real time as the news comes out, this is a pretty interesting workflow. I’ll leave a link in the description if you want to check out Co-Invest. And now, let’s get back into the video. But another really interesting use case that you can do with Jev is now you can use Jev here to

06:11 control your browser because yes you can be to control your browser with AI but it’s a lot slower right if you’re using Jack GBT browser use or computer use or if you’re using claude it consumes a lot more tokens it consumes a lot of time right so with Jeff and browser use you can see the browser user now also integrated Jeff inside of it so if you have to use it you also have to join the weight list I already added myself to the weight list so by the time you watch my next video you should to see me reviewing browser use with Jeff as well to see how fast it really is. Okay, but

06:42 you can see that from this video here, someone has take a source and a destination and Jeff here is able to use browser use and be able to book a flight, right? Be able to navigate and search flights here under 10 seconds. You can see that how much it took and essentially what it does here is basically try to feed in all the operations that browsers can do to Jeff. And Jeff here is going to make a decision on exactly where to click, what to do, right on each and every single page. And of course, it requires LM here to feed in exactly what are the options.

07:13 And Jeff here is going to basically try to get the response a lot faster. But however though, if you actually want to use Jev right away, you can see the versel actually have Jeff here inside of AI gateway. So if you click on the models here and just going to search for Jeff for example, and here you can see that Jeff here is completely for free. You can try it out inside of Versell and currently you can see it released that in September 15. And let’s say if you actually want to use Verscell here for Jeff, you can simply just use their AI SDK and is able to just call it directly and get the decisions or you can also

07:43 use a curl request here to call it and it’s going to also help you to do that. Now, on top of that, Jeff here also released their claw skills or AI agent skills, which essentially what you can do now is that you can be able to pass a skill here to cla or any other AI agents. And that skill repository will basically teach your AI agents on how to use jab and how you can have jab here to making small decisions inside of your applications or inside of your day-to-day AI automation workflows. So, pretty much that’s it for this video. I’ll make sure to put all the resource that I talked about in this video in the link to description below so you can actually be able to copy it and be able

08:14 to use it right away. So, like I said, I’ll make sure to give Jeff here a try. And if you want to see me make another video on Jeff and also comment down below what kind of demos you want me to make for Jeff. So, please comment down below for that and I’ll make sure to make a video on that very, very soon. So, with that being said, that’s it for this video. I’ll see you in the next video.