The industry has evolved thanks to increasingly larger models and more powerful computer chips, particularly Nvidia’s Graphics Processing Units (GPUs). However, the most advanced models are often restricted from China. Two contrasting points have emerged in this conversation. One is that DeepSeek has managed to build an incredibly robust Large Language Model (LLM) using a Mixture of Experts (MoE) architecture with much less computing power, electrical consumption, and time than traditional models.
Let me take a step back to explain MoE, using a healthcare analogy to highlight why it represents a truly innovative way to use AI.
What is a Mixture of Experts (MoE)?
An MoE model is like having a team of specialists, each with a specific area of expertise, working together to solve a problem. Instead of relying on a single large model to tackle everything, MoE divides the task among different “experts,” and a smart system known as a gating network decides which experts to consult for each particular input.
Imagine you walk into a hospital with a health concern. Rather than seeing a general physician for everything, you might be directed to a specialist—say a cardiologist for heart issues or a neurologist for brain-related concerns. The triage nurse, in this case, acts as the gating network, deciding which expert (doctor) is best suited to handle your problem.
In AI, the experts are smaller neural networks trained to focus on different aspects of a problem. The gating network acts as a decision-maker, choosing which experts should address a given input. The selected experts then contribute to the final output, often with their contributions weighted based on their relevance to the task at hand.
This is a simplified explanation of how MoE works. For a more technical look, here’s an example from Mistral AI:
Why is MoE Efficient?
MoE is highly efficient because it activates only a few experts at a time, significantly reducing computational power and energy usage. It scales well, allowing large AI models to tackle complex tasks more effectively. By distributing work among specialized experts, MoE can process vast amounts of data while ensuring quick and accurate responses. This approach is used in large-scale AI applications, like Google’s NLP models. But what sets DeepSeek apart is its fundamental reliance on MoE for its speed, efficiency, and cost-effectiveness.
What Else Makes DeepSeek Special?
In addition to its innovative use of MoE, DeepSeek made its entire core programming software open-source. You can download a lighter version for your desktop, and if you have a high-end gaming computer (perhaps one with Nvidia’s RTX 4090, retailing for around $1,600), you can run the full platform. While your local setup may not have the processing power or training that DeepSeek’s cloud-based application offers, you can still get into the AI game for less than $10,000.
DeepSeek chose to license its software under the MIT License model, which is simple, permissive, and collaborative. Below is the entire user agreement:
MIT License
AI and Consumer Health
There’s a clear, shared obsession with improving consumer health, especially considering the struggles facing the $5.2 trillion U.S. healthcare industry. Issues such as staffing shortages, rising operating costs, reduced reimbursements, and deferred care from the COVID-19 pandemic continue to plague the sector, causing quality to lag behind other developed nations. On the other hand, AI has the potential to improve the consumer experience dramatically. While it’s not yet ready for widespread daily use in medicine, we see its potential growing, and reports about it are increasing.
For example, a study published on January 20, 2025, in the Journal of the American Medical Association (JAMA) tested an AI tool designed to draft responses to patient messages. The tool, called PAM Chat, was deployed across nine clinics, with 166 healthcare professionals using nearly 2,600 AI-generated drafts out of over 21,000 suggestions. Interestingly, 88% of these AI-generated responses were completely rejected by the physicians. While the article presents the 12% success rate as a positive outcome, it’s important to note that these responses were not accepted wholesale. Rather, the doctors only used parts of the AI’s suggestions.
This shows that AI still has a long way to go before it can match the consistency and accuracy of a skilled physician. We believe that, in order for AI to be widely adopted by healthcare providers, it must meet three progressive benchmarks. The first and most critical is that it must be as reliable as doctors—every time. We can’t move on to the next steps until this foundational goal is met.
That said, we remain optimistic. With access to consumers’ longitudinal healthcare records, AI has the potential to transform interactions between patients, their providers, and even payors (particularly when dealing with denied services). We can make healthcare faster, better, and more cost-effective than ever before, and we believe this can be achieved today.
About HealthScoreAI ™
Healthcare is at a tipping point, and HealthScoreAI is positioning to revolutionize the industry by giving consumers control over their health data and unlocking its immense value. U.S. healthcare annual spending has exceeded $5 trillion with little improvement in outcomes. Despite advances, technology has failed to reduce costs or improve care. Meanwhile, 3,000 exabytes of consumer health data remain trapped in fragmented USA systems, leaving consumers and doctors without a complete picture of care.
HealthScoreAI seeks to provide a unique solution, acting as a data surrogate for consumers and offering an unbiased holistic view of their health. By monetizing de-identified data, HealthScoreAI seeks to share revenue with consumers, potentially creating a new $100 billion market value opportunity. With near-universal EHR adoption in the USA, and advances in technology, now is the perfect time to capitalize on the data available, practical use of AI and the empowering of consumers, in particular the 13,000 tech savvy baby boomers turning 65 every single day and entering the Medicare system for the first time. Our team, with deep healthcare and tech expertise, holds U.S. patents and a proven track record of scaling companies and leading them to IPO.
Noel J. Guillama-Alvarez
https://www.linkedin.com/in/nguillama/
+1-561-904-9477, Ext 355
https://en.wikipedia.org/wiki/MIT_License
https://github.com/doxdk/deepseek-desktop?tab=readme-ov-file
https://www.proxpc.com/blogs/gpu-hardware-requirements-guide-for-deepseek-models-in-2025
https://jamanetwork.com/journals/jama/fullarticle/2829691