Solo Coder Says a Big AI Lab Copied His Free Idea
He gave away his AI design a year ago. A big lab just got credit for it.
A lone developer has posted a frustrated note saying he built and gave away an AI design a year ago, and that a big AI lab recently announced something very similar as a major breakthrough. In March 2025 he published a research paper, a working model, a training dataset and a free software package, all openly available. A year later, he says, a frontier lab showed up with the same core idea, but without publishing a paper, without releasing the model, and without sharing the data behind it.
So what does the technology actually do? Most AI today writes and thinks one word at a time, like typing a sentence letter by letter. His approach skips that, predicting many things at once to produce fast, structured answers. In plain terms, it takes a back-and-forth conversation and quickly spits out a number from 0 to 1, a probability, for each turn: how likely is this chat to turn into a sale? He built it for sales conversations specifically.
The two systems are similar, not identical. His version trains the AI by rewarding good guesses, using something called reinforcement learning (teaching by trial, reward and error). The lab's version, called Jev, samples many answers in parallel and produces confidence scores instead of a single number. Still, the developer says the core architecture and the vertical use case line up closely. His broader complaint is emotional and familiar: months of hard work, lost sleep, and almost no recognition, because a big lab made a general-purpose version that grabbed all the attention.
The bigger lesson is about credit in open-source AI. Independent builders publish freely, then watch larger, better-funded labs get the headlines for similar work. That does not prove anyone copied anything, and the two designs may differ in important ways. But it does show how much of today's AI progress quietly builds on unpaid, unheralded work by individuals. Everything he made is still free to download and inspect, which is more than can be said for the lab's version.
- A one-person developer says he published and gave away this AI design in March 2025, including the paper, the model and the training data
- Big labs tend to get noticed for broad, general tools, while niche work by independent coders often goes unseen
- His system predicts sales outcomes as a score from 0 to 1 for each turn of a conversation, quickly and without the usual word-by-word approach
Why It Matters
Big labs often get credit and money for ideas independent coders already shared for free.