San Francisco-based Mirendil Startup has attracted $200 million in funding, reaching a valuation of $1 billion and becoming a "unicorn." Investments were made by venture giants Andreessen Horowitz (a16z) and Kleiner Perkins, as well as the chip manufacturer Nvidia. The company, founded by former employees of Anthropic Behnam Najshaber and Harsh Mehta, intends to focus on creating artificial intelligence systems capable of recursive self-improvement, but with one important caveat - these technologies will be available to open-source developers.

Self-improving AI is considered by many experts to be both a potential key to solving global problems such as climate change and treating major diseases, and a source of serious risk. While industry leaders, including Anthropic and OpenAI, are calling for global committees to oversee such technologies, Mirendil offers an alternative path. The company believes that the problem is not the most recursive improvement, but its monopolization of a handful of closed laboratories.

Today, any scientific group trying to use AI in drug development, chemistry, or robotics must become a parallel AI laboratory. Our goal is to democratize access to advanced AI tools , - claim the startup’s founders.

The choice of a company’s name, referring to the Elvish language from "The Lord of the Rings," looks ironic against the backdrop of heated arguments about AI safety. Yet Mirendil’s ambition is serious: it wants to place powerful algorithms in the hands of independent scientists to accelerate scientific progress. Amid recent incidents of forced disablement of powerful models at the behest of authorities and calls for tough regulation, Mirendil’s move looks like a bold bet on decentralization and openness to scientific discovery.