Launching svGPT, plus AI in Healthcare

Launching svGPT, plus AI in Healthcare
Today we are launching our first AI service: svGPT, a large language model-based chatbot—for paid members only—trained on a massive corpus of sports medicine, wellness, and general health information. You can also read the FAQ for more.

If you're a paid subscriber you can try it now. If you're not a paid subscriber and you'd like to try it, then you need to upgrade to a paid plan.

This is only an early marker of our plans. So, if you're interested, you can read on to find out more about how we think about AI in healthcare, where we're going, and where we're not.

Here is a short demo:



We are quietly optimistic about the potential for AI services in healthcare, training, fitness, and general wellness. There is huge potential to bring new ideas and approaches to people via AI, if we avoid falling into many traps, some obvious, others less so. We have plans for various new services, and we are launching the first one today: svGPT.

Before explaining what it does, it is important to go deeper into our thinking. We don't want to fall into the trap of Dr. Google, where people search massive data stores for symptoms, assume the worst, and then panic or schedule expensive testing. When you hear hoofbeats think horses, not zebras, as the saying goes.

Humans are not algorithms

But we should want to be careful about how we apply AI to human health. People are not algorithmic, nor are they anarchic. We don't need more scanning, testing, and treatment—we generally need less. And statistical patterns of the sort current large language models rely upon only get you so far. Humans are messy baskets of biology, sprouting with cognitive biases, and both more fragile and more resilient than they seem. Caution is not just warranted, but required. Most people need less customization in improving their health, not more. Attempts to get gains with more specialized approaches generally go badly and expensively wrong.

These new models and approaches are important, however. Much of medicine is about language, hence how clinicians use the term "presents", as in how does someone with a problem present: What words do they use? How do they sound? What is the message in what they say about how they feel? Medicine and language are closely intertwined, so anything that helps us untangle human presentation and see through it to underlying causes is important, and potentially life-changing.

How can AI help? While AI has long existed in medicine, it has generally been more oriented toward probabilistic reason around data, like whether someone in an ICU surrounded by instruments, blinking lights, and machines that go ping, might be about to crash, to use the medical term for having things go very, very badly.

The average physician stops listening in less than 15 seconds

What is new, however, is that technology can finally move past symptoms and penetrate the veil of language itself. What do we mean when we mean? What is inside the language when a patient presents, often rambling and unintentionally obscurantist? Studies show the average doctor stops listening within 15 seconds and has a fixed idea of what's going on. That's often efficient, but far from always. Large language models don't get bored and stop listening.

Equally important is how these tools allow rapid compilation and synthesis of vast corpora (data sets) of research findings, clinical outcomes, experience, and heuristics. This is the other side of language, the kind of language with which medicine talks to itself. But the flood of such language is overwhelming, a cacophony impossible to capture and process, unless you are a language-based model that doesn't stop listening.

Can't current large language models do this? Sort of, but not really. These models have, for practical purposes, ingested the entire public internet. This drives these massive LLMs to mediocrity in a similar way to how a mutual fund that owns all stocks must, by definition, eventually underperform the market by exactly the amount of its fees. A large enough LLM trends toward aggregated mediocrity.

Our thinking is different. We have taken the approach of training svGPT on the latest research and clinical data, all carefully focused on sports medicine, wellness, fitness, and nutrition. Gigabytes and gigabytes of targeted data to the exclusion of all else. In a perfect world, our chatbot would refuse all other questions (and sometimes it does).

So this is where we stand, quietly optimistic but careful. We are not making grand claims, nor are we going to stand aside and pretend that penetrating language itself doesn't have consequences for sports medicine, wellness, nutrition, etc.

That is why we are now launching our first service, an AI chatbot trained on an extensive and targeted corpus in sports medicine and wellness. It is, unavoidably, for paid subscribers only, given its resource requirements, and it is the first of various practical AI services we have planned, all in the service of simplification.

svGPT is now available, but for paid subscribers only. You can find it here. It can also be accessed through our new AI item on the main navigation bar. There is also an associated FAQ. Keep an eye out for more AI-related tools and services.

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to Simplavida.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.