The Kids’ (Notes) Are Alright

Volume 25 September 15, 2025

From the Editor

Good morning readers, we hope you had a great weekend!

Take Surya Dantuluri’s piece on The Training Imperative: the argument is simple, if you don’t train your own model, you’ll always be stuck with generic results. That resonates with PT. Our field is built on tailoring care to the individual, and I think AI for PT will follow the same rule. The tools that matter won’t be the off-the-shelf chatbots; they’ll be the ones shaped by rehab data, adherence patterns, and real-world clinic workflows.

Then there’s the pediatric SOAP-note study. The fact that AI-generated notes are holding their own against humans is less about replacement and more about relief. If AI can take the first pass at structured documentation, that frees us up for the things machines can’t do, picking up on subtle cues, motivating kids, adapting care in real time.

Finally, it’s encouraging to see PT education catching up. The Hawai‘i Pacific University study shows students are not just using AI but learning how to question it. That’s the kind of mindset that will make AI an ally instead of a crutch.

My read? AI in PT is moving forward, but now it’s time to take it from the general to the specialized. PTs shouldn’t be using programs designed for the masses—we should be getting programs tailored to us. We tailor our treatments to patients, so why not have a system that actually tailors to us?

📰 This Week's Highlights

1. The “Training Imperative”: Why Apps Must Own Their Models to Drive Real Value

📖 Read the full article → (6 min)

Surya Dantuluri argues that AI app builders must move beyond simply using lab-provided APIs—eventually, they must train their own models. He traces the shift from early LLM access via APIs to today’s growing feasibility of model distillation, fine-tuning, and custom training. DeepSeek distilled a frontier model for just $6 million in early 2025—illustrating the drop in barriers to entry. The real moat? Token Factor Productivity (TFP)—economic value delivered per token.

Why this matters for PTs:
Owning data from patient interactions—exercise adherence, documentation quirks, even how we cue patients—could let PT apps fine-tune models specifically for rehab. Thinking in terms of “value per action” may help clinics evaluate ROI on AI tools.

2. AI vs. Human SOAP Notes: Pediatric Rehab Study Finds Surprising Parity

📖 Read the preprint → (8 min)

A 2025 study compared AI-generated SOAP notes with those written by clinicians in pediatric rehabilitation. Using two different generative AI tools, researchers evaluated content quality, accuracy, and clinical completeness. Findings showed AI-authored notes were often comparable to human ones, with strengths in structure and clarity—but sometimes weaker in nuanced observations. Human reviewers rated many AI notes as acceptable for clinical use, though supervision remained essential.

Why this matters for PTs:
This research suggests AI could shoulder some documentation load without sacrificing quality, especially for routine or structured notes. For PTs in pediatric or high-volume settings, AI support may free bandwidth for patient engagement while still requiring clinician oversight.

3. AI in PT Education: Integrating Tools Builds Confidence and Research Skills

📖 Read the narrative review → (10 min)
📖 Read the Hawai‘i Pacific study → (7 min)

Recent studies underscore AI’s role in PT education. One review calls for intentional adoption of AI tools in DPT programs to boost access, productivity, and feedback—while noting concerns about misinformation and plagiarism. At Hawai‘i Pacific University, AI tasks—like article searching and abstract writing—were integrated into an Evidence-Based Practice course. Surveys showed significant gains in students’ AI literacy and research self-efficacy.

Why this matters for PTs:
AI literacy is becoming a core skill. Exposing PT students to AI early may improve evidence-based practice and prepare graduates for AI-augmented clinical environments

Bonus Reads

• 🖨️ Personalized Rehab via 3D Printing + AI – Researchers are combining AI and 3D printing to create custom rehab devices tailored to a patient’s unique anatomy and needs.
  📖 Read the study → (7 min)

• 🧠 AI in Neurorehab: Fact vs Fiction – A thoughtful review on what AI can really do in neurorehabilitation, and where the hype runs ahead of reality.
  📖 Read the review → (10 min)