News Summary
TREE NEWS reports: On August 23, TechCrunch reported that Inherent, a London-based AI lab founded by former Google DeepMind researchers, has unveiled Faraday, an AI research agent designed to independently replicate results from published scientific papers. In benchmark tests, Faraday outperformed frontier models such as Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, despite being built on a relatively small Qwen 3.6 model with only 2.7 billion parameters.
Industry Analysis
Faraday’s success challenges the prevailing assumption that larger models are inherently more capable. By leveraging reinforcement learning to cultivate ‘research taste’—the ability to judge which experiments are worth pursuing and how to design them—Inherent is pushing AI beyond simple pattern recognition into hypothesis-driven scientific inquiry. This approach mirrors how human researchers are trained: replication of existing work as a foundation for original discovery.
The fact that Faraday uses OpenAI’s GPT-5.5 Codex as a coding tool highlights a pragmatic, tool-using paradigm. Rather than relying solely on its own capabilities, Faraday acts as an orchestrator, calling external models for specialized tasks—much like a scientist using software to analyze data. This hybrid workflow could become the norm for AI agents in research, enabling smaller, more efficient models to punch above their weight.
Inherent’s seed funding of $50 million and its ambition to build an ‘AI scientist’ capable of autonomous discovery across multiple domains signal a new frontier in AI research. However, the company’s focus on validation through replication is a critical first step, ensuring that AI-generated findings are reliable before they are trusted to generate novel knowledge.
Forward-Looking Perspective
If Inherent’s training methodology proves scalable, we could see a shift in how AI models are developed for scientific applications—prioritizing efficiency and domain expertise over raw parameter count. This could democratize access to advanced research tools, allowing smaller labs and institutions to compete with tech giants. For the broader AI and crypto ecosystem, the integration of AI agents with decentralized compute networks or data marketplaces could unlock new use cases, such as on-chain peer-reviewed research or tokenized scientific discoveries.
As Faraday evolves, the key metric will be whether it can move from replication to genuine innovation—producing hypotheses that lead to new, verifiable scientific breakthroughs. If successful, Inherent may not only redefine AI’s role in science but also set a precedent for how AI and blockchain technologies can collaborate in the pursuit of knowledge.
