Yes, Jev Is Insane, But There's A Catch
Read the Portuguese translation →
Author: Two Minute Papers · Watch on YouTube
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: Yes, Jev Is Insane, But There’s A Catch · Channel: Two Minute Papers · 04:50 · Watch on YouTube
Topics: AI agents, Model inference and optimization, AI models, Local and open-source models
Summary
Section titled “Summary”The video explains that Jev is designed to make decisions by choosing among options rather than generating text, and that its speed could help with real-time tasks and simple decisions in AI workflows. It describes three elements of Jev: a decision-focused architecture, parallel sampling of possible answers, and an RLCD training method intended to make confidence reflect accuracy. The video notes that related ideas have been explored in earlier research, while presenting Jev as a useful combination of existing and new ingredients. Its verdict is that Jev is promising, but also amplified by hype and viral marketing.
Key points
Section titled “Key points”- Jev selects an answer from a question and a set of options instead of writing text.
- The system is described as producing decisions up to about 200 times faster than current chatbots.
- Its speed could enable real-time video game play and help with simple decisions in AI systems.
- Delegating small choices to Jev may speed up some AI workflows and reduce token usage.
- The video notes that no official research paper is available, only a high-level website description.
- The basic idea is not entirely new, and earlier open research has explored similar approaches.
- Jev combines a decision-focused architecture, parallel sampling, and a training method called RLCD.
- RLCD is intended to align the system’s confidence with how often it is correct.
- The video calls Jev a useful tool while cautioning that hype and viral marketing have amplified attention.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | What Jev does | Jev chooses among possible answers rather than generating text, and its speed could support real-time decisions. |
| 01:04 | Using Jev in AI workflows | The video describes offloading simple decisions to Jev and raises the question of how its approach relates to older classifiers. |
| 02:10 | Earlier research and Jev’s ingredients | The video acknowledges similar earlier work and outlines Jev’s architecture, parallel sampling, and RLCD training. |
| 03:14 | Confidence and the verdict | Confidence estimates could guide when to trust Jev or defer to a heavier model, and the video weighs its promise against hype. |
| 04:18 | Closing and Lambda promotion | The presenter explains the time spent preparing the video, invites viewers to subscribe, and promotes Lambda’s AI tools. |
Quotes
Section titled “Quotes”Instead, it makes decisions. — 00:00
And if it is extremely fast, it can play things in real time. — 00:32
They call it parallel sampling. — 02:44
if it says 80% confident, it should be correct about 80 times out of 100. — 03:14
Is it also heavily amplified by hype and viral marketing? Yes. — 04:18
Concepts and entities
Section titled “Concepts and entities”Jev, classifier, decision model, parallel sampling, RLCD, Lambda, NVIDIA
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 A new AI called Jeff is taking the world by storm, but why get excited for an AI system that can’t even write text? It properly ran over the internet, so let’s try to separate the facts from the hype. Okay, so it does not write text. Instead, it makes decisions. A question and some options go in, one answer comes out. Kind of like Who Wants to Be a Millionaire, but you don’t wait for [music] the answer, it pops out instantly, up to about 200 times faster
00:32 than our chatbots today. That is very useful. You can also play video games with it. Hooray! The question is the world state and the potential answers are the moves you can make. And if it is extremely fast, it can play things in real time. But it gets better. LLMs are relatively slow, so when you code up something with a super smart AI, everything takes a long time. But along the way, you also have a few simple decisions too. Which file do
01:04 I open? Which failure matters? Retry or move on? And this is where Jeff can help. Offload these decisions to Jeff and it solves them nearly instantly. In some cases, it can meaningfully speed up a fable-like AI system and even cut down the token usage. It makes things cheaper and faster. That is amazing. Now, there is no research paper, but some of you fellow scholars probably say, “Okay, this is a classifier, an idea that dates
01:37 back around 90 years now. So, what’s new here?” Well, first, there is no official research paper. There is only a high-level description on the website, but it doesn’t stop us fellow scholars from doing our research, does it? Now, hold on to your papers, fellow scholars. Come on, this time with more feel. >> [music]
02:10 [music] [music] >> Now we’re talking. Actual song, by the way. So, similar ideas have been explored for years and years now. There are open papers surprisingly close in spirit, and we celebrate amazing science and research here. So, I would like to say a
02:44 big thank you to the scientists behind these earlier open works. So, the basic idea in Jev is certainly not new, but Jev also combined three ingredients. One, a new architecture that is built to make decisions instead of generating tokens. Two, it evaluates all possible answers all at once instead of one by one. They call it parallel sampling. And three, a new training method they call
03:14 RLCD, where the AI’s confidence roughly matches how often it is right. So, if [clears throat] it says 80% confident, it should be correct about 80 times out of 100. Not always. Why is that good? Well, this way you can actually look at the confidence value and decide when to trust it or fall back to a heavier, smarter model. Once again, similar ideas have been explored for years now. But, the way this one is put together, it is an incredible
03:48 combination that gives us a really useful new tool. Love it. And you fellow scholars are also pumping out open source implementations of it as well. That is open science at its best. You are amazing. So, let’s be level-headed and separate the facts from the hype. Is it awesome? Yes. One ingredient closer to fast local AI that we will be able to own forever. Now, is it also heavily
04:18 amplified by hype and viral marketing? Yes. Is it new? Partly. Okay, I took a bit longer to make this for you. That’s not great for views, but I do it to be more concise, look up related papers, and see how you fellow scholars use it in practice. That is a better video for you. Subscribe and hit the bell for more papers. I use Lambda to reproduce AI research papers often in minutes. It’s also great to train your own models or find tool an
04:50 existing one. or text-to-image or video, easy peasy. Running a deep seek chatbot or agent, super fast, super reliable. Lambda gives you powerful Nvidia GPUs to run your own experiments. I test ideas from the papers I cover and moments later, results. Love it. Seriously, try it out now at lambda.ai/papers.