A journalist at WIRED built PitchFly, an article idea generator based on the open-source drosophila connectome
A journalist at WIRED created PitchFly, a project that uses the freely available drosophila brain connectome (166 000 neurons, 125 million synapses) to generate ideas for article headlines – without understanding the meaning of words, simply by remixing learned patterns.
A journalist at WIRED described how he created a project called PitchFly. It uses an openly published connectome – a detailed map of the brain of a male drosophila (fruit fly), developed by researchers from Google and a number of academic institutions. The map captures how 165 000 to 166 000 neurons connected by 125 million synapses communicate with one another, and makes it possible to simulate how a fly would respond to various stimuli.
For PitchFly, the author collected hundreds of the most successful article headlines from the WIRED website over the past year and used Codex to convert them into a representation a neural network could process. These data were then "fed" into the connectome, which generates its own headline suggestions based on them. According to the author, this is not a language model – the fly brain does not understand the meaning of words; it merely rearranges and remixes learned patterns in new ways.
According to the article, several dozen other projects using the same fly brain model emerged after the connectome was published in early September 2025. These include training it to play Beat Saber, a trading bot called StonkFly created by an engineer at Coinbase (according to the author, it loses money, but the results are surprisingly decent given the approach), and attempts to teach the connectome to play Doom, Minecraft, and Pong, solve a Rubik’s Cube, or drive a virtual car. The author points out that he has not verified whether some of these projects actually use the connectome, and that some may merely be animations or jokes; alongside them, however, there are also serious scientific tools for visualizing and manipulating neural circuits.
According to the author, mapping the brains of different animals offers a path toward understanding the biological foundations of intelligence – the human brain, with its 86 billion neurons, cannot yet be mapped, but smaller creatures may provide useful insights. The author also notes that the ease of training and deploying such a small, specialized model suggests that a major benefit of large language models may lie precisely in building smaller, specialized tools for specific tasks.
Why it matters
The project is experimental and playful, but it shows that openly published biological data can easily be turned into a working machine learning demonstration using common code generation tools, without having to train a large model of your own. It also demonstrates growing interest in small, narrowly focused models as an alternative to general-purpose large language models for specific tasks.
Relevant practical impact
What this means
For individuals
It shows that freely available datasets (such as the drosophila connectome) and tools like Codex allow an individual without extensive resources to quickly build their own experimental AI tool for a specific recreational or work task.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.