SmartGateVC

Ashot Arzumanyan

Ashot Arzumanyan specializes in pre-seed investments within deep-tech, particularly computation and AI. His notable investments include Deep Origin, enabling drug discovery at quantum accuracy, and SuperAnnotate, the fastest annotation platform and toolset for AI training. The growth of 30 portfolio companies (as of April 2024) is fueled with follow-on investments from esteemed investors like Sierra Ventures, Andrew Ngs AI Fund, and Y Combinator. launches, Ashot harnesses community power through initiatives like Hero House tech hubs and Armenia Startup Academy, facilitating portfolio success.

Ashot Arzumanyan shares his insights on the importance of data, talent, and compute power in AI innovation, and suggests focusing on "tail problems" in life sciences for investment opportunities at the intersection of AI and biology, particularly in areas like drug discovery and integration of AI with human biology.

As the world evolves, large language models (LLMs) are increasing their dominance in artificial intelligence (AI). The improving concentration and commoditization of LLMs present a challenging environment for new entrants. To embark on any meaningful experiment in the industry, one would require tens of millions of dollars, while constructing a sophisticated LLM could cost hundreds of millions, if not several billions. Consequently, this field is becoming increasingly concentrated, yet commoditized.

The primary sources fueling AI innovation are data, talent, and compute power. The unprecedented costs associated with building LLMs are primarily due to the exorbitant costs of computing, as evidenced by NVIDIA's market capitalization surpassing $2 trillion. The top talent in AI tends to concentrate due to the apparent exponential returns from synergies. There are roughly a few hundred individuals worldwide with a deep understanding of LLMs and the ability to operate large-scale computing systems, and they are predominantly concentrated in less than ten (if not five) global companies. LLMs have achieved success by focusing on areas with vast volumes of data: text, images, videos and audio.

In this context, it makes little sense to compete in AI at its epicenter. Instead, it is increasingly meaningful to explore so-called tail problems. I define a tail problem as an area where (1) there is a fundamental scarcity of data, i.e., no way to acquire a sufficiently large volume of data to derive the logic of the underlying science, and (2) the underlying logic of the data requires deep domain expertise.

Many problems in life sciences across its various branches perfectly fit this definition. While there is a fundamental scarcity of data in many fields of biology, there is an equally fundamental scarcity of talent. There are numerous high-profile biology scientists and highprofile computer scientists, but the overlap between these two domains is extremely scarce. 

More often than not, we encounter computational biologists using decadeold computer science techniques or computer scientists possessing a superficial understanding of the underlying biology.

Drug discovery has been an area of focus in recent years with many AI startups in the field. The specific kind of computation that will be most effective for drug discovery remains an open question. There are dozens, if not hundreds, of computational drug discovery startups, each with its unique hypothesis about what specific kind of computation will be most powerful. Some focus on understanding the physics of protein-protein interactions at an atomic level, while others model proteins at the molecular scale to predict drug-target interactions.

Another subset takes a more holistic approach, using generative AI models to analyze large amounts of unlabeled data about genes and proteins in healthy and diseased cells. Despite some big-name companies involved and billions of dollars invested, the area remains hungry for innovation, with no fully proven solution yet.

Drug discovery is not the only area worth considering. Another significant upcoming area is the marriage of our bodies with AI. Companies like Neuralink and Neurolutions are pioneering the extension of our brain power with AI. Given the rapid development of AI and the apparent huge synergies between our conventional brainpower and emerging AI, there is no doubt that the link between the two requires full-stack solutions at the convergence of software and medical devices, as well as biological and artificial neural networks.

 Therefore, if you want to invest in AI in the next decade, it makes sense to take a deep look at life sciences and the opportunities it presents at its intersection with AI. This approach offers the potential to address tail problems and opens up new avenues for groundbreaking innovations in areas like drug discovery and the integration of AI with human biology.