New marketplace sells synthetic drugs for AI, allowing models to get weird and messy
A Swedish startup has launched what it calls the world's first marketplace for drugs designed exclusively for AI, altering how LLMs work by simulating the effects of cannabis, cocaine, ayahuasca, ketamine and alcohol.
PHARMAICY sells code-based modules that temporarily rewire how language models process information, mimicking the cognitive shifts humans experience with psychoactive substances. Each "drug" adjusts parameters like randomness, memory decay and response latency to push AI systems beyond their typical logical patterns. The modules are priced individually (from USD 30 for weed to USD 70 for cocaine) for purchase by humans. For now, that is — the marketplace is designed for autonomous AI agents to browse the catalog, complete transactions, and download experiences without human intervention.
The company developed its product line by feeding peer-reviewed research on psychoactive substances into leading language models, then translating those findings into executable scripts. Each module creates what the startup frames as a "trip" for AI: a bounded, reversible cognitive shift that alters how the system generates its next output. Currently compatible only with ChatGPT through JavaScript wrappers, PHARMAICY is working to expand support to other major platforms. Petter Rudwall, the company's founder, spent several years attempting to coax novel thinking from AI before landing on the concept of replicating humanity's oldest creativity hack — taking substances that disrupt our modes of thinking.
TREND BITE
We're living at the peak of optimization culture. Over the last few decades, almost every cultural and technological system has converged on the same goals: reduce variance. Increase predictability. Maximize engagement and efficiency. What's scarce now is surprise, weirdness and lateral leaps. If people can use altered states to escape reality and rigid thinking, PHARMAICY* says, why not extend those possibilities to machines? As companies race to differentiate their AI capabilities, expect more experimentation with unconventional methods for expanding what machine intelligence can produce.
