@archiexzzz: I trained the fly to become a gymbro - to do bicep curls and also do squats for a good leg day. Mapped all 139,255 proo…

X AI KOLs Following News

Summary

The user trained a fly model to perform gym exercises like bicep curls and squats using the DeepSeek-V4-Flash AI model and Cline Desktop app, mapping 139,255 neurons from the FlyWire connectome on a MuJoCo physics engine.

I trained the fly to become a gymbro - to do bicep curls and also do squats for a good leg day. Mapped all 139,255 proofread neurons, 50M+ synapses, and neurotransmitters. Everything was done using the open-source DeepSeek-V4-Flash model on the @cline 's Desktop app, an open-source app for open-weight models that i had early beta access to. 68 bodies / 102–103 joints / 78 actuators - flybody as the base model running on the MuJoCo physics engine. FlyWire v783 connectome: 139,255 proofread neurons, 130M synapses (Zenodo record 10676866, CC-BY 4.0) flywire_annotations supplemental TSV -> 118,104 static soma coordinates rendered (4×4×40 nm voxels → µm) Reading layer: pandas / PyArrow (feather chunked access paths), requests (resumable Zenodo downloader) Cline is very easy to use and required very little effort to make this work. It did a lot of web searching and correctly compiled everything with very little token usage.
Original Article
View Cached Full Text

Cached at: 09/14/26, 11:33 PM

I trained the fly to become a gymbro - to do bicep curls and also do squats for a good leg day.

Mapped all 139,255 proofread neurons, 50M+ synapses, and neurotransmitters. Everything was done using the open-source DeepSeek-V4-Flash model on the @cline ’s Desktop app, an open-source app for open-weight models that i had early beta access to.

68 bodies / 102–103 joints / 78 actuators - flybody as the base model running on the MuJoCo physics engine.

FlyWire v783 connectome: 139,255 proofread neurons, 130M synapses (Zenodo record 10676866, CC-BY 4.0) flywire_annotations supplemental TSV -> 118,104 static soma coordinates rendered (4×4×40 nm voxels → µm)

Reading layer: pandas / PyArrow (feather chunked access paths), requests (resumable Zenodo downloader)

Cline is very easy to use and required very little effort to make this work. It did a lot of web searching and correctly compiled everything with very little token usage.

Similar Articles

I Trained a Fly’s Brain to Generate WIRED Story Ideas

Wired

Using a fruit fly's brain connectome, a detailed neural map, researchers and hobbyists have created AI applications like generating story ideas and trading stocks, showcasing innovative and humorous uses of biological intelligence.