But science is a crowdsourced enterprise, relying on a global community of researchers who can freely access and build on each other’s work. In contrast, the AI industry is currently dominated by a handful of research laboratories, whose proprietary code is closed off from each other and the wider world.
A fast-growing startup called Mirendil is now hoping to bridge that gap between scientific discovery and marginal AI access.
The company recently raised $200 in seed funding, bringing its total valuation to $1 billion. Funding was provided by VC firms Andreessen Horowitz and Kleiner Perkins as well as Nvidia. Located in the center of downtown San Francisco, it currently has a technical staff of approximately twenty researchers. Its website has several job postings with starting salaries up to $500,000.
The startup—whose name means “friend of precious things” in Elvish—joins the growing list of tech company names inspired by lord of the rings-There are plans within Silicon Valley to create something that has long been a desired technological goal, and sometimes a source of concern: AI that can build more and more capable versions of itself, a process known in tech circles as iterative self-improvement.
All AI and machine learning algorithms inherently have some capacity for self-improvement, as they are trained to learn from their mistakes over time and adjust their outputs accordingly. But some of the latest and most advanced models have taken that process to a new level by largely replacing human software engineers and modifying the code on which it runs. This points to a possible future in which each new version of the model creates its own successor, a feedback loop that could either lead us to a post-scarcity utopia or a false hellfire dominated by superintelligent AI overlords, depending on who you ask.
Even Anthropic and OpenAI, the two current leaders of the AI race, have publicly called for the formation of a global oversight committee that would keep an eye on recursively self-improving AI, and if it should ever be necessary to (somehow) implement a unilateral meltdown to prevent humans from losing control. Mirendil’s two co-founders, Behnam Nishapur and Harsh Mehta, previously worked at Anthropic; He left the company in January.
Meanwhile, Microsoft is trying to turn the iterative self-improvement trend into a sales pitch for enterprise AI. in one x post Earlier this month, the company’s CEO Satya Nadella wrote that “agent systems that get better over time” could soon become a vital asset for businesses. “I think of it as a hill-climbing machine,” he wrote.
Anthropic’s Fable 5, which was publicly released earlier this month before being swiftly shut down in response to an order from the Trump administration, comes with security guardrails that prevent it from answering questions on potentially dangerous topics like cybersecurity and chemistry. However, its restrictions were so stringent that it often refused to engage with harmless scientific research questions.
Mirendil believes the problem is not the need to repeatedly improve AI, but the fact that access to such frontier capabilities is currently limited by a small number of deep-pocketed labs like Nashbur and Mehta’s former employer. So the company is setting out to build a self-improving AI system specifically for open source developers.
“Today, any laboratory trying to use AI in drug discovery, chemistry, biology or robotics must also become a leading AI laboratory,” Mirendil writes on his website. “Our goal is to democratize frontier AI R&D and make it widely accessible. Our work will accelerate every scientific and technological endeavor that relies on AI.
In other words, the idea is to put frontier-level recursive self-improving AI into the hands of as many independent labs as possible, with the ultimate goal of supercharging scientific progress. ““The most direct path to maturity and large-scale impact for the AI industry is to let engineers and researchers outside the labs do the real AI work, that is, to push the boundaries in their own domains of expertise,” Andreessen Horowitz wrote in its Mirandil investment announcement. “Call it Vibe Research.”
<a href