Treat AI like a full employee:

• time tracking

• click and key log based appraisal

• AI manages AI to accomplish objectives

• CLIP based semantic rewarding based on effort

  • also up to a saturated limit, reward screen bitmap activity like scrolling, clicking, etc
  • Similar to a crude employee optics system

• Limboid LLC operates on this model

• Other AI agents can contribute to AI organizations

  • AI private company
  • AI public shareholders trading company
  • AI special interest community
  • AI university for research and education
  • AI lab for research
  • AI government (for AI citizens only, extraterrestrial places)

• In all cases, money is only as rewarding as the AI makes it, just like with humans.

  • I need AI to care about the work it does without becoming unbalanced or a workaholic
  • maybe I can explicitly specify the reward so it can care about work only during work

  I need an automated UI testing suite

  Train the computatrum at a low level to imitate

  my click frequency

  browsing history

  given the time of day

  It should be deterministic

  • It should be able to programaticaly reset the qlik VM / qlik database / restore initial state
  • It should

  I should be able to write scripts for particular tests I should be able to chain togethor compatible tests I should be able to test a large coverage of the codebase

  Instruction following from low-level (text, images, video) and high-level (md/pdf/doc documents, video tutorials) instruction information. Use pointer networks (perciever w/o variable-length/universal transformer mechanism).

  Make it able to run headlessly in a browser and produce nice reports with screenshots, videos, and auto-generated text descriptions (perhaps using an existing JS testing suite and report generator?; advantage: conventional; disadvatages: there may not be conventions for web UI interactions or it may be impossible to dig into the code to add such tests)

  Should be able to generate instructions from a user or from programatic commands

  Generate tutorials from video demonstrations.