JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Chinese field notes

Day-one notes from a skeptic, in Chinese

Jiayuan Zhang argues that Jev looks like a faster general classifier, then narrows its plausible advantage to finite action spaces with tight latency requirements such as browser control, games, and robotics.

Original by Jiayuan ZhangAgent workflowsX postOriginal published Source reviewed

Before you dive in

What you’ll find in the original

  1. A finite DOM action set fits a Jev loop, but long-context browser quality and reasoning still need benchmarks.
  2. Many tool calls cannot be replaced because their arguments—such as exact edit lines—must be generated rather than selected from a fixed menu.
  3. Turn-based games, robotics, and autonomous systems offer bounded decision trees where latency may matter, though an LLM can still perform some of these tasks.
Worth knowing

These are one day of exploratory notes. The author had not seen benchmark evidence and describes the poker and Pokémon systems as demos still in progress.

More guides →See the experiments →