Human brain cells do next-token prediction | Parasma

We're now backed by Y Combinator. Read the note →← NewsSAN FRANCISCO — August 13, 2026 — Parasma today announced its public launch. The company is developing the algorithms and infrastructure for compute powered by human brain cells as an energy efficient alternative to GPUs.

The work behind Parasma first entered public view earlier this year. In a project unveiled in late February, founder Sean Cole collaborated with Cortical Labs to train a culture of human neurons to play Doom.

Token prediction

The architecture converts context words into electrical stimulation delivered to the cells. Their responses are measured and decoded directly into tokens, then the system feeds back whether each prediction was correct.

Parasma tested whether human brain cells can solve precisely defined token prediction tasks. The cells correctly predicted 90% of next-token decisions while taking in the full sentence context. Randomized control runs scored significantly worse, with all results consistent with chance.

Parasma also ran a harder, non-linear token prediction experiment where no single context token alone can determine the answer. In one isolated run, the cells achieved 78.1% accuracy , exceeding the 75% ceiling that a silicon linear decoder could achieve.

“Crossing this boundary that a linear silicon decoder cannot represent is exactly the kind of result we built Parasma to pursue,” said Sean Cole, founder of Parasma . “It is an early step toward understanding where living neurons can add computational capabilities, particularly in places like language modelling.”

These results suggest that custom architectures built for biological computing can support language-based token prediction.

What this could enable

In the near term, Parasma is building toward repeatable benchmarks across cultures, devices, and time to scale biology with advances in AI. The company will also investigate richer contextual prediction and adaptive-control tasks.

Over the longer term, programmable biological compute would replace silicon in systems that benefit from continual learning or extreme energy and sample efficiency, giving rise to new model paradigms. Parasma’s goal is to develop the algorithms that make biological computing useful as an alternative to GPUs.

Keep reading<br>View all

Parasma is now backed by Y Combinator<br>Company18 Jun 2026