Jev: Ultra-Fast AI for Decision-Making

Jev represents a new class of AI, purpose-built for rapid decision-making at scale, capable of processing hundreds of thousands of decisions nearly instantly.

The system excels when provided with clear, structured context and specific decision criteria, evaluating inputs against defined logic to output structured results. Matthew Berman emphasized that its primary distinction is its ability to "make thousands, tens of thousands, hundreds of thousands of decisions incredibly quickly and incredibly inexpensively."

Users define decision logic by structuring questions and criteria in JSON format, which Jev then processes.

For instance, to determine if a hot dog is a sandwich, users must define 'sandwich' and 'not a sandwich' criteria within the JSON structure. Berman illustrated this, stating, "You have questions and you're structuring this as JSON. So the instructions or basically the prompt is is hot dog a sandwich?" The model resolves these queries by assigning probability percentages, such as '73% true' for specific questions.

Jev is not a generative model, meaning it is unsuitable for unstructured tasks like providing advice or writing code.

Instead, it is particularly useful for dynamic UI assembly, selecting from existing buttons, fonts, and assets to construct web pages.

The model outperforms traditional Large Language Models (LLMs) such as Fable in time-constrained scenarios due to its instant decision speed, making it highly effective for specific, rapid-response applications. As Berman explained, "It's not a text generation model. You're not going to ask it an unstructured question expecting an answer."

Jev functions as an 'AI slop detector,' capable of auditing websites like entropic.com to identify AI-generated content, thereby aiding in content moderation.

Kitsy's 'Unclutter' tool leverages Jev to remove ads, cookies, and pop-ups from web pages in real-time, significantly improving the user browsing experience. Matthew Berman noted, "It is quite a good slop detector. So, if you think that a website or some kind of creative writing has been written by AI, you can actually have Jev check it for you."

The pricing for Jev's services is highly competitive, costing 4.2 cents per million input tokens, with output tokens provided for free.

This open-source tool offers extremely fast results, enabling users to type in general descriptions and have the system highlight exact matches on the page, as Berman articulated: "You just type in whatever you want to find on the page, it doesn't have to be exact, you just have to type in kind of what you're looking for, and it will highlight on the page exactly where to find it."

Jev does not generate code from scratch but dynamically assembles webpage components by selecting from a preconfigured library of UI elements.

The system efficiently renders forms or sign-in blocks in real-time based on user intent, demonstrating its high speed in interface construction. Berman clarified that "it's not building it from scratch, it's just choosing from a bunch of different UI elements and putting them all together in the right way."

Clipfast, an application utilizing Jev, scans video transcripts to identify and clip relevant segments from lengthy videos.

Users can input specific topics to instantly generate multiple clips from a 90-minute video in under two seconds, complete with one-click download functionality, showcasing Jev's speed. Matthew Berman provided an example: "This is an example of Jev clipping a 90-plus minute video into multiple clips based on whatever you type in in under 2 seconds."

Jev exhibits a sophisticated contextual understanding by mapping world knowledge entities to specific color palettes, such as those associated with '80s disco' or characters like 'Mario.' Matthew Berman explained, "So, it knows what colors are associated with each of these world knowledge entities." Additionally, it performs contextual emoji selection based on user text input, with new users receiving $5 in initial credits to begin using Jev's capabilities.