How to Use Jev for Fast, Structured Decisions

TL;DR

Jev makes large numbers of structured decisions quickly and inexpensively when given clear criteria, context, and JSON-formatted questions. It is best suited to classification, ranking, selection, and matching tasks, including filtering website clutter, prioritizing emails, locating relevant passages, assembling interfaces from existing components, and finding video clips, rather than generating prose or writing code from scratch.

Transcript

Jev is everywhere and it is the most exciting AI that I have seen in a long time. It really does feel like the early days of chat GPT where we had this thing that was so powerful and so raw and people are trying to figure out how to get the most out of it. And so I'm going to try to explain today not only what is Jev, but also how to use it, how to... Read More

Key Insights

• Jev is a decision-making AI that can evaluate thousands, tens of thousands, or hundreds of thousands of choices quickly and inexpensively. Its value comes from applying defined criteria repeatedly at a scale and speed that would be difficult for slower AI models or manual workflows.
• Jev works best when a task contains clear context, structured questions, explicit criteria, and known possible outcomes. The sandwich example uses JSON definitions for true and false classifications, then returns a probability and resolves the question according to the stronger result.
• Jev is not designed primarily for open-ended text generation or code writing. It is poorly suited to unstructured requests such as asking for workplace advice, but it can choose among existing components, categories, actions, or other predefined options with high speed.
• Website clutter removal works by sending page elements to Jev with instructions describing what counts as spammy or annoying. Jev classifies advertisements, cookie banners, upsells, and unwanted dialogs so the application can remove those elements almost immediately after the page loads.
• Workflow automation can use Jev as a decision step between connected applications. The Zapier example proposes evaluating each new calendar invitation against user-defined criteria, then automatically deciding whether the invitation should be accepted or declined within a workflow spanning supported applications.
• Email prioritization is a strong fit because Jev can rank messages by expected response priority instead of relying on chronological order. One demonstration sorts 100 emails in under half a second, showing how rapid evaluation can make a large inbox more actionable.
• Dynamic interface assembly is possible when Jev receives a preconfigured library of buttons, input fields, fonts, forms, and other UI assets. It selects and arranges appropriate elements for a user's needs, but it does not generate the underlying interface components from scratch.
• Video clipping works by scanning a long video's content, probably through its transcript, and identifying segments relevant to a typed request. The Clipfast demonstration processes a video longer than 90 minutes and produces multiple matching clips in under two seconds for one-click downloading.
• More videos from this channel:
  • How to Use Codex for Faster, Smarter Workflows
  • How fast will AI self improvement change humanity?
  • How to use Claude Co-work for non coding tasks
• More from this channel: Matthew Berman

Questions & Answers

Q: What is Jev and how does it make decisions?

Jev is described as an artificial intelligence focused on decision-making rather than open-ended content generation. A user supplies a structured question, relevant context, possible classifications, and explicit criteria describing how each outcome should be recognized. Jev evaluates the input against those definitions and can return a probability, as shown when the sandwich example received a 73 percent true result.

Q: When is Jev a good fit for an AI task?

Jev is a good fit when the desired task can be expressed as a large set of structured choices governed by clear rules. It is especially useful for classification, ranking, selection, and matching because it can make thousands, tens of thousands, or hundreds of thousands of decisions quickly and inexpensively. Clear context, structured questions, and well-defined criteria are central requirements for strong results.

Q: What tasks is Jev not good at performing?

Jev is not intended to serve as a general text-generation model that answers loosely framed questions or produces long, open-ended responses. The transcript specifically says it is unsuitable for asking advice about how to speak with a boss and is not very good at writing code. Its strengths appear when it chooses among structured options rather than creating unrestricted prose or software from scratch.

Q: How can Jev remove ads and clutter from websites?

Jev can receive the elements found on a webpage along with a description of how to identify content that is spammy, distracting, or annoying. It then classifies items such as advertisements, cookie banners, upsells, and unwanted dialogs so a browser tool can remove them. The demonstrated Unclutter tool performs this cleaning nearly instantly and allows users to provide their own key.

Q: How can Jev improve email and calendar workflows?

Jev can evaluate email or calendar data against user-defined priorities and then trigger an appropriate action. For email, it can sort messages by how important it believes a response is, with one demonstration ordering 100 emails in under half a second. In Zapier, it can assess each new calendar invitation and automatically choose whether to accept or decline it based on supplied criteria.

Q: How does Jev enable smarter find-in-page search?

Traditional find-in-page tools depend on exact keyword matches, so they fail when the requested wording does not appear literally on the page. The demonstrated fuzzy-search tool lets a user describe what they want to locate without entering the exact text. Jev identifies the relevant content quickly and highlights where it appears, providing a more flexible alternative to strict browser keyword matching.

Q: Can Jev build a website or user interface?

Jev can assemble a webpage dynamically when it has access to a preconfigured library of interface assets, including buttons, input fields, fonts, forms, and sign-in elements. It decides which components suit the user's needs and arranges them rapidly. However, it is not building the website from scratch or writing all the code itself. Its role is selecting and combining existing components.

Q: How can Jev find clips in a long video?

Jev can scan the content of a long video, probably by evaluating its transcript, and match relevant segments to a user's typed request. In the Clipfast demonstration, it searches a video longer than 90 minutes, identifies multiple suitable clips, and completes the process in under two seconds. The resulting clips can then be downloaded with one click, making rapid content selection practical.

Summary & Key Takeaways

• Jev is presented as a decision-making AI designed to evaluate structured questions against explicit criteria. Its central advantage is the ability to make thousands, tens of thousands, or hundreds of thousands of decisions quickly and inexpensively. It works best when developers clearly define the question, relevant context, possible outcomes, and standards for choosing among them.

• The demonstrated applications focus on classification, ranking, matching, and selection. Jev can detect AI-like content, remove unwanted webpage elements, route automation events, prioritize emails, perform semantic find-in-page searches, locate relevant video clips, generate associated color palettes, and select emojis that match a phrase or described intent.

• Jev is not positioned as a general text generator or strong code-writing model. Its interface-building capability depends on choosing and arranging components from a preconfigured library rather than producing an application from scratch. The examples show that its value comes from combining structured options, explicit decision rules, low cost, and very fast processing.