Soon the fishbrain will be frontrunning you in the trenches
$zneuro wil beat $flybrain
The fly walked so that the fish could run
80kmc. Early.
$anthropig $catgpt $jubjub $frontier
The fish meets Robinhood.
We’re planning Fish Market Lab: historical data → paper trading → a wallet with strict limits.
Replay the same market, remove one input, compare the trades. Test what drives the model before putting real funds behind it.
11
643
$zNeuro 10m coin larping at 100k right now
The fish is cooking. This is even bigger than $flybrain
$catgpt $anthropig $moo $boner
What is the model actually doing on Wikipedia?
It receives screenshots, extracts visual contrast and turns that input into cursor movement, permitted link clicks and scrolling. The live view shows the visual input, the model’s activity and the resulting browser actions.
Wikipedia gives us a real environment for testing this loop. The controller currently responds to visual patterns; it does not understand the articles, and its parameters are fixed.
Alongside it, the Target test lets us measure specific tasks: reaching a target, responding when it moves and continuing when visual input is briefly interrupted.
The longer-term direction is to develop and train a more capable agent, then evaluate it on tasks it has not seen during training.
One proposed environment is historical market replay: learning a decision policy, testing it on unseen periods and recording simulated buy, sell and hold decisions. If those evaluations justify it, a later stage could connect the agent to a wallet with spending limits and separate execution controls.
That is the research path from moving a cursor toward making more useful decisions.
Watch the experiments:
zebraneural.com
Roadmap:
zebraneural.com/docs.html#ro…
8
863