Video summary
why EVERYONE is copying this
Main summary
Key takeaways
Summary
The video introduces Jev, a decision-focused AI model from TypeSafe. Unlike ChatGPT or Claude, which generate natural-language responses, Jev returns structured decisions, such as a yes/no answer with confidence, a multiple-choice selection, or a score.
The presenter describes Jev as a “System 1” model, borrowing Daniel Kahneman’s term for fast, intuitive judgments. He compares it to an AI-powered if statement or logic gate.
Jev is designed to work inside software and automated workflows, rather than primarily as a chatbot or a source of open-ended advice. Its narrow, fast outputs may make it useful for routine classification and routing tasks. However, Jev’s weights and training data are closed, so its underlying advantages are not fully known.
Tests and Comparisons
- In a spam-comment example, Jev returned a confident classification almost immediately. In one single-query comparison, the presenter reported roughly 2 seconds for ChatGPT versus 17 seconds for Jev, though results will depend on the models and setup.
- The video also processes a batch of 100 channel comments, highlighting speed and cost as potential advantages for handling many decisions.
- In a later comparison of Jev, OpenAI’s Decisions API, and local decision models—including Clef, Nimble, and Tev—Jev performed best overall in the presenter’s test. OpenAI’s API came close, while Clef 27B was the strongest local model.
- The presenter said the cloud models were generally more accurate and that OpenAI’s Decisions API cost about 2.4 times as much as Jev in this test.
Demonstrations and Use Cases
- Model routing: Jev quickly chooses a small, medium, or large model based on a prompt’s difficulty, reserving more capable models for harder questions.
- “Intelliprompter”: A teleprompter transcribes the presenter’s speech with ElevenLabs and sends it to Jev every half-second. Jev checks whether each planned talking point has been covered, including when the presenter paraphrases it.
- Browser automation: Jev helps navigate Domino’s ordering flow. The presenter reports completing an order in about 15 seconds, compared with about 61 seconds using Codex.
- Other projects: JevGrep helps coding agents find relevant files; its creators claim it can reduce agent costs by 40%. The video also mentions a Hermes plugin that delegates yes/no decisions to Jev and a Doom-playing demonstration.
The presenter frames decision models as a potentially useful new building block for developers and people using AI coding agents. He emphasizes that they complement, rather than replace, general-purpose language models.
Sponsor Mention
The video includes a Twingate sponsorship. The presenter describes Twingate as providing secure remote access without opening ports, with access scoped to specific people, devices, or applications. He demonstrates a service account that lets a cloud-based AI agent access one selected home-network app rather than the rest of the network.
Main Speakers and Sources
- NetworkChuck — Presenter and narrator; demonstrates the tools and reports the comparison results.
- Diogo Almeida — TypeSafe cofounder; discusses Jev’s purpose and decision models.
- Ali — TypeSafe team member; discusses how developers and coding agents might use decision models and shares the Intelliprompter idea.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.