Fordel StudiosInception raises $50 million to build diffusion models for code and text
What Happened
Diffusion models already power AI image generators, but Inception thinks they can be even more powerful applied in software development.
Our Take
$50M to apply diffusion models to code is confidence bordering on delusion. Look, diffusion works beautifully for images because humans have visual intuition—we can see when something's off. Code? Syntax is brittle. You can't "interpolate" between two functions and get something useful.
The real challenge isn't the model, it's that code has hard constraints. Diffusion's strength is handling ambiguity in continuous spaces. Programming is discrete. They're solving the wrong problem with the right technology.
What To Do
Skip following diffusion-for-code closely until they ship a public demo—sentiment alone shouldn't convince you this actually works.
Cited By
React