bitcoin
bitcoin

$63555.814907 USD

-1.77%

ethereum
ethereum

$1802.843574 USD

-0.63%

tether
tether

$0.998885 USD

0.00%

bnb
bnb

$588.197072 USD

-0.70%

usd-coin
usd-coin

$0.999842 USD

-0.01%

xrp
xrp

$0.980457 USD

-1.10%

solana
solana

$74.831635 USD

-0.32%

tron
tron

$0.312645 USD

0.97%

hyperliquid
hyperliquid

$68.082158 USD

-1.31%

zcash
zcash

$936.887928 USD

-1.89%

monero
monero

$502.129266 USD

-0.75%

dogecoin
dogecoin

$0.059341 USD

-0.29%

unus-sed-leo
unus-sed-leo

$9.342086 USD

0.55%

chainlink
chainlink

$9.150014 USD

-1.30%

cardano
cardano

$0.153592 USD

-0.35%

Cryptocurrency News Video

What is Jev? Help AI Agent save decision-making costs|Ansys error message classification test|What Is Jev? Cheaper AI Agent Decisions

Nov 04, 2026 at 10:29 am Jeff Hong AI Simulation Lab

Tools & projects: https://github.com/ovo0615 The narration of this film is an AI-synthesized voice, the images are animated, and there is no screen recording. The classification result of Jev is taken from a request actually sent on 2026-11-03 (jev-1.13.0); the original text of the error message is taken from the execution record of the previous AI Agent modeling video, and the host name and project name have been obscured; about 10 seconds from 00:33 are the original footage of that video. Prices are publicly quoted in October 2026. The narration is AI-generated and the video is animated, not a screen recording. Jev's results come from one real request sent on 2026-11-03 (jev-1.13.0); the error messages come from the run log of our earlier AI agent model-repair video, with host and project names masked; about 10 seconds from 00:33 are footage from that video. Prices are public prices list as of October 2026. AI Agent modifies models, builds models, and runs solutions. Every step he takes costs money. When encountering a multiple-choice question like "What kind of question is this?", you can first hand it to Jev: a model from TypeSafe Company. Give it a paragraph of text and several options, and it will choose one and give each option a probability. The answer is in a fixed format, and the program can directly go to the next step. ・Relay: When Agent encounters an error message, it first blocks the path and host name, and then hands it to Jev. If the confidence level is 0.80 or above, it will be dealt with according to that category; if it is less than 0.80, it will be returned to the Agent for detailed inspection, or handed over to the engineer. ・Real run: Five error messages that were actually encountered in the modeling video, sent to Jev in one request. 1. Two parts interfere with each other: choose geometry problem, confidence 1.00; Agent later did return to Discovery to repair the interference before solving it. 2. Failed to generate grid: judged as "unknown", probability 0.79. You can't tell the reason just by looking at this sentence. The real reason is the interference in Article 1. 3. The host name cannot be found: the highest selection connection (0.48), the confidence is only 0.42; the actual reason is the setting of distributed solution, ranking third. 4, 5. The confidence is less than 0.80, so send it back for closer inspection. Only one of the five is certain; if it is not certain, it will tell the truth. ・Price (per million input tokens, USD): Jev is about 0.042, output is free; Claude Haiku 5.5 is 0.10 (about 2.4 times); GPT-5.6 Luna is 0.20 (about 4.8 times); Claude Sonnet 5.5 is 2 (about 48 times, output is excluded). ・The five messages this time used 2,280 input tokens. Same run a thousand times: Jev about $0.10, Sonnet 5.5 light input about $4.56. Jev is only responsible for classification and route selection; it is still up to the Agent and the engineer to decide whether to modify the model and deliver it. Related videos|Previous videos of AI Agent model modification: https://youtu.be/ylEkGnc25eQ A few easy-to-read mistakes: ・Five messages and one request are a small sample test and do not represent the accuracy of Jev on all error messages. ・"Confidence" and "probability" are two numbers: probability is the score Jev gives to each option, and confidence depends on whether the probability is concentrated in one option. ・The "actual reason" in the film is based on the execution record of the Agent, not the output of Jev. ・0.80 is the threshold for customizing the process of this film, not the built-in value of Jev. ・The cost of one thousand times is calculated and only counts input; the output of other models is charged separately. ・The fees and token counters at the beginning are indicative values. Introduction to Jev and TypeSafe: https://docs.typesafe.ai/introduction Jev, TypeSafe, Claude, GPT, Ansys, HFSS, and Discovery are trademarks of their respective companies. This channel has no affiliation or endorsement relationship with them. This video is not an official tutorial from any manufacturer. This video does not contain customer information. Technical communication: jeff.hong@cadmen.com This channel is Jeff Hong's personal technology sharing and is not the official account of Ansys, TypeSafe or any AI manufacturer. Every step an AI agent thinks through, while repairing models, building them or running solves, costs money. Questions like "what kind of problem is this?" are multiple choice, and that is what Jev is for: a model from TypeSafe that takes some text and a few options, picks one and gives each option a probability, in a fixed format that code can act on directly. - The relay: when the agent hits an error, it masks paths and host names and passes the message to Jev. At 0.80 confidence or more, the agent handles it as that category; otherwise it goes back to the agent for a closer look, or to an engineer. - Real run: five real error messages from the model-repair video, sent to Jev in one request. Two parts intersecting: geometry, confidence 1.00, and the agent did fix it in Discovery. Mesh generation failed: "unknown" at 0.79; the real cause was the interference. Host name not resolved: connection at 0.48 with confidence 0.42; the real cause, a distributed-solve setting, ranked third. The other two also stayed under 0.80. Jev was sure about one of five, and said so when it was not. - Price per million input tokens (USD): Jev about 0.042 with free output; Claude Haiku 5.5 0.10 (about 2.4x); GPT-5.6 Luna 0.20 (about 4.8x); Claude Sonnet 5.5 2 (about 48x, output billed on top). - The five messages used 2,280 input tokens. A thousand runs: Jev about 0.10 USD; Sonnet 5.5 about 4.56 USD for input alone. Jev only classifies and picks a path; repairing the model and deciding to deliver stay with the agent and the engineer. Earlier AI agent model-repair video: https://youtu.be/ylEkGnc25eQ Notes: five messages in one request is a small trial, not an accuracy figure. Probability is Jev's score per option; confidence reflects how concentrated those scores are. The "actual cause" comes from the agent's run log, not from Jev. The 0.80 threshold is a demo value. The thousand-run cost is an input-only estimate. The opening cost and token counters are illustrative. About Jev and TypeSafe: https://docs.typesafe.ai/introduction Technical collaboration: jeff.hong@cadmen.com Chapters 00:00 Why: every agent step costs money 00:33 The earlier model-repair video of AI Agent 00:49 What is Jev: multiple choice + confidence | What Jev is: multiple choice plus confidence 01:24 The relay: agent, Jev, agent 01:48 Five real error messages, one real run 03:08 How much cheaper: price per million input tokens 04:02 Who does what, and recap #AIAgent #Ansys #LLM
Video source:Youtube

Disclaimer:info@kdj.com

The information provided is not trading advice. kdj.com does not assume any responsibility for any investments made based on the information provided in this article. Cryptocurrencies are highly volatile and it is highly recommended that you invest with caution after thorough research!

If you believe that the content used on this website infringes your copyright, please contact us immediately (info@kdj.com) and we will delete it promptly.

Other videos published on Nov 20, 2026