Reactive AI
Reactive AI team found a way to make AI 25x cheaper by storing all context inside the model and looping it to self-improve toward superintelligence. It achieves 77% of SOTA performance today with as little as $100K in compute. The latest generation is up 27% and closing fast. He's backed by EWOR Fellowship, Snowflake founder, inventor of AI reasoning at OpenAI, a 6x unicorn founder, and angels backing Magic dev, FigureAI, and CuspAI.
About the Tech
Large Language Models were built to respond, not to become superintelligent. They are amnesic by design and can't truly improve themselves without backpropagation. Current AI models can do the work. This leaves the AI services market, about 18× larger than model APIs, and structurally out of reach for today’s models. Reactive AI is bringing cognition and execution from the agent layer into the model itself, unlocking a market opportunity far larger than LLMs.
About the Team
Wiktor was the only bachelor’s student in the top 1000 Innovators @ Stanford among 230 PhDs. He scaled his first AI startup to 180K users before age 18. Adam is a self-taught researcher who after reading 60 AI/ML books built an alternative AI architecture to GPT in 2024. Weronika won Intel AI Global out of 1000+, ex-Founder in AI/ML pre-GPT, and R&D team leader of 10 at University of Warsaw and TGYD@Tsinghua.
More from EWOR
