--- title: Fruit fly connectome repurposed for stock trading,... description: Hobbyists are repurposing the fully mapped fruit fly (Drosophila melanogaster) connectome — about 160,000 neurons and 25 million synaptic edges — for tasks far... canonical: https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Fruit fly connectome repurposed for stock trading, games, and OCR | daily.dev og:description: Hobbyists are repurposing the fully mapped fruit fly (Drosophila melanogaster) connectome — about 160,000 neurons and 25 million synaptic edges — for tasks far... og:url: https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej og:image: https://api.daily.dev/og/posts/oMAPy56ej.png og:image:alt: Fruit fly connectome repurposed for stock trading, games, and OCR og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Fruit fly connectome repurposed for stock trading, games, and OCR **[Collections](https://daily.dev/sources/collections)** · 2 min read · 3 upvotes · 0 comments ## Summary Hobbyists are repurposing the fully mapped fruit fly (Drosophila melanogaster) connectome — about 160,000 neurons and 25 million synaptic edges — for tasks far outside biology. Projects include Stonkfly, a stock-trading bot built on the connectome's reward circuits, and attempts to control games like DOOM, Mario 64, and Beat Saber via simulated dopamine-driven reinforcement. The newest, FlyOCR by Jerry Liu, uses the MaleCNS v1.0 connectome to read PDFs by mapping pixel glyphs to receptor activations and running current-based neuron dynamics, hitting roughly 87% accuracy on a Microsoft 10-K filing — comparable to older MNIST-era models but behind modern document parsers. The interest lies less in raw benchmark scores than in using an actual biological wiring diagram, rather than gradient descent, as a computational substrate. ## Content Researchers recently completed mapping the full central nervous system connectome of both male and female *Drosophila melanogaster* — the common fruit fly — charting roughly 160,000 neurons and 25 million synaptic edges. The maps revealed sexually dimorphic circuits tied to mating behavior, but hobbyists have since found stranger uses for them. One project, Stonkfly, wires the connectome's reward circuits into a stock-trading bot. Others have fed simulated sensory input through dopamine-driven reinforcement to get the connectome controlling games: DOOM, Mario 64, and Beat Saber have all been attempted. The latest experiment is FlyOCR, built by Jerry Liu. It uses the full MaleCNS v1.0 connectome to read PDFs. The pipeline splits a PDF image into individual glyphs, maps pixels to receptor activations, runs simplified current-based dynamics across all 166,000 neurons and 25 million edges, then applies a compact readout model on the resulting spikes. Tested on a Microsoft 10-K filing, it hit around 86% accuracy on balance sheet headings and handled numeric values reasonably well. Across 1,700+ sampled characters and digits, overall accuracy came in at 87% — comparable to older MNIST models, though well short of modern document parsers. What makes these projects interesting isn't the benchmark numbers. The wiring comes from an actual animal nervous system rather than gradient descent on training data. Whether that distinction matters for performance is unclear, but it's a genuinely different starting point — and people keep finding new things to plug it into. ## Questions this post answers ### What accuracy did the fruit fly connectome achieve when used for OCR on a real document? Around 87% overall accuracy across more than 1,700 sampled characters and digits, tested on a Microsoft 10-K filing, including about 86% accuracy specifically on balance sheet headings. This is comparable to older MNIST-era digit recognition models but well short of modern document parsing systems. The project, called FlyOCR, ran the full MaleCNS v1.0 connectome (166,000 neurons, 25 million edges) with simplified current-based neuron dynamics and a compact readout model on top of the resulting spikes. _Curious how biologically-inspired computing stacks up against trained models? Follow experiments like this on daily.dev._ ### How many neurons and synapses are in the fully mapped fruit fly connectome? The complete central nervous system connectome of Drosophila melanogaster contains roughly 160,000 neurons and 25 million synaptic edges, mapped for both male and female flies. The maps revealed sexually dimorphic circuits linked to mating behavior, and the wiring has since been repurposed by hobbyists for tasks like stock trading bots, game-playing agents, and optical character recognition. _Track unconventional applications of biological neural wiring in computing on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 28 comments across x (as of 2026-09-14). **TL;DR:** Reactions are mostly amused fascination at using a real fly connectome for OCR, tempered by pointed technical pushback that the demo's real-world robustness and comparative performance are weaker than the framing suggests. **Sentiment:** 30% positive · 45% mixed · 25% skeptical **The case for** - Novel use of an actual biological wiring diagram rather than gradient descent is seen as genuinely interesting and a sign of real experimentation. - Some see the 87% accuracy on real financial glyphs as a meaningful signal given the constraints of the substrate. **The pushback** - Segmentation reportedly fails with just a 3° tilt, and successful table reads rely on pre-supplied row/column geometry rather than robust parsing. - A raw-pixel linear classifier reportedly scores 99% versus the circuit's 89% on digits, and only 1 of 8 PDF lines was exactly correct. - Degree-preserving rewiring of the connectome reportedly retains most of the accuracy, raising doubts about how much the specific fly wiring actually contributes. - Skepticism that this is more rebranding of familiar components under an "AI" label than a fundamentally new capability. **By community** - x (mixed): A wave of hype and jokes about the fly-brain-reads-PDFs concept sits alongside a substantive critical thread questioning whether the connectome itself is doing meaningful work or if the demo is cherry-picked and fragile. **Hottest debate:** Whether the fly connectome's specific wiring is actually responsible for the OCR results, given claims that degree-preserving rewiring retains most of the accuracy and a simple linear classifier outperforms it on digits. **Open questions** - Does the receptor-activation approach generalize to noisy, skewed, real-world scans rather than clean or pre-cropped glyphs? - Is the segmentation failure caused by the tilt itself or by an assumption of axis-aligned rows in the cropping step? - How much of the OCR performance is attributable to the fly-specific connectome versus generic network structure (per the proposed degree-preserving rewiring control)? **Highlights** > @jerryjliu0 @nftechie_ cool but the post omits that a raw-pixel linear classifier scores 99% against the circuit's 89% on digits, that only 1 of 8 PDF lines is exactly right, and that degree-preserving rewiring retains most of the accuracy not to be that guy, i'm massive fan of this arch, which is > — [\_DMontgomery40 on x · 1 points](https://x.com/_DMontgomery40/status/2099416408275001854) > @byzaidrais @jerryjliu0 @nftechie_ The README reports a segmentation failure with just a 3° tilt. The successful tables are selected crops with rows and columns supplied by page geometry: > — [ShuohaoLiao on x · 1 comments](https://x.com/ShuohaoLiao/status/2099422843389829603) > @jerryjliu0 @nftechie_ A useful control: shuffle the connections while preserving neuron degrees, then train the same readout. That could help show how much the fly wiring contributes to OCR. > — [aethmere on x · 2 points](https://x.com/aethmere/status/2099169744423440866) > @ShuohaoLiao @jerryjliu0 @nftechie_ Fair catch, 3° shouldn't break segmentation if the geometry pass is doing its job. Worth checking whether the failure is the tilt itself or the crop assuming axis-aligned rows to begin with. > — [byzaidrais on x](https://x.com/byzaidrais/status/2099475776307368243) > @jerryjliu0 @nftechie_ This feels like a good example of when the "AI" part just means calling a well-known component a few new things. > — [first\_sauce\_lab on x](https://x.com/first_sauce_lab/status/2099200409692766375) **Source threads** - [x](https://x.com/jerryjliu0/status/2099168936059437206) · 0 points · 28 comments ## Similar posts on daily.dev - [Second complete map of a fruit fly brain completed](https://daily.dev/posts/second-complete-map-of-a-fruit-fly-brain-completed-kabloxiit) · Ars Technica · 0 upvotes · 0 comments - [Someone wired up a digital fly's brain in a Pi 5 to play Pokémon, and you can help it progress](https://daily.dev/posts/someone-wired-up-a-digital-fly-s-brain-in-a-pi-5-to-play-pok-mon-and-you-can-help-it-progress-rdqqak8o5) · XDA Developers · 0 upvotes · 0 comments - [Deep-learning model predicts how fruit flies form, cell by cell](https://daily.dev/posts/deep-learning-model-predicts-how-fruit-flies-form-cell-by-cell-fhntelpcm) · MIT News · 90 upvotes · 7 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#reinforcement-learning](https://daily.dev/tags/reinforcement-learning) [View this post on daily.dev](https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Fruit fly connectome repurposed for stock trading, games, and OCR","url":"https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej"},"datePublished":"2026-09-14T15:38:26.233Z","dateModified":"2026-09-14T15:39:15.160Z","description":"Hobbyists are repurposing the fully mapped fruit fly (Drosophila melanogaster) connectome — about 160,000 neurons and 25 million synaptic edges — for tasks far...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/14c74fa8e615be814eb7691f816bf2c9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/14c74fa8e615be814eb7691f816bf2c9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,reinforcement-learning","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Fruit fly connectome repurposed for stock trading, games, and OCR"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/fruit-fly-connectome-repurposed-for-stock-trading-games-and-ocr-omapy56ej#faq","mainEntity":[{"@type":"Question","name":"What accuracy did the fruit fly connectome achieve when used for OCR on a real document?","acceptedAnswer":{"@type":"Answer","text":"Around 87% overall accuracy across more than 1,700 sampled characters and digits, tested on a Microsoft 10-K filing, including about 86% accuracy specifically on balance sheet headings. This is comparable to older MNIST-era digit recognition models but well short of modern document parsing systems. The project, called FlyOCR, ran the full MaleCNS v1.0 connectome (166,000 neurons, 25 million edges) with simplified current-based neuron dynamics and a compact readout model on top of the resulting spikes. Curious how biologically-inspired computing stacks up against trained models? Follow experiments like this on daily.dev."}},{"@type":"Question","name":"How many neurons and synapses are in the fully mapped fruit fly connectome?","acceptedAnswer":{"@type":"Answer","text":"The complete central nervous system connectome of Drosophila melanogaster contains roughly 160,000 neurons and 25 million synaptic edges, mapped for both male and female flies. The maps revealed sexually dimorphic circuits linked to mating behavior, and the wiring has since been repurposed by hobbyists for tasks like stock trading bots, game-playing agents, and optical character recognition. Track unconventional applications of biological neural wiring in computing on daily.dev."}}]} ```