I had two unusually interesting experiences with AI at Work this week. In one, Chat GPT diagnosed a quite arcane computer software problem, involving a commerical site, my lap top, and my router.
In the other, I asked Chat GPT to take a kind of weird unexpected position on the value of Point of Care Testing, and it came back with some interesting ideas.
### SOFTWARE DEBUGGING
Header: Chat GPT fixed a complicated multi-part software problem that sidelined my work for half a day.
Nowadays, we rely heavily on cloud-based software—Google Drive, for example. I use a database and information-management system called Notion. For a day or two, it began performing terribly: web pages refused to open or took up to a minute to load. Even after simple pages finally appeared, a progress indicator kept spinning, as if the system were still trying to download something.
My first assumption was a (rare) server-side problem at Notion. But their status page showed everything was up and running. ChatGPT suggested a simple test: run my desktop computer using my iPhone’s Wi-Fi hotspot instead of my home network. When I did, Notion worked dramatically better.
At that point, ChatGPT concluded that the problem wasn’t Notion, and it wasn’t my computer—it was my router.
After some back-and-forth troubleshooting, we disabled the 6 GHz band on my router (which supports 4/5/6 GHz). ChatGPT explained that this also disables experimental router software associated with Wi-Fi 7. Once we made that change, Notion immediately returned to normal performance.
The solution sounds straightforward when explained this way, but the path to it was anything but obvious. I never would have guessed that a hard-wired Ethernet desktop could be disrupted by a Wi-Fi setting—especially one (6 GHz) I hadn’t even known existed.
(Boring, but long dialog here.)
### NEW IDEAS IN POCT
Header: When asked to provide an unusual frame and outlook for the value of point of care testing, I thought Chat CPT came up with an interesting and clever approach.
##
I was reading several recent review articles on point-of-care testing (POCT) when a stray thought crossed my mind: Shannon information theory. I have only a superficial understanding of the theory, but out of curiosity I gave ChatGPT a set of current POCT review articles and asked whether it could identify any new or interesting value implications by viewing POCT through that theoretical lens.
What came back was, at minimum, genuinely interesting. ChatGPT reframed POCT not simply as a logistical or workflow innovation, but as an information-processing system, and explored how concepts from information theory might help explain observed advantages of POCT in clinical practice.
To my eye, this was a good example of AI doing something more than summarization or pattern matching—it was generating novel perspectives and structured ideas in response to an open-ended question.
Important caveat: I am not claiming that this constitutes publishable, ground-breaking research, or that it should form the basis of a PhD thesis. Rather, it struck me as a thoughtful and somewhat surprising example of AI producing genuinely interesting conceptual work when prompted in the right way.
For more detail of what it actually said, here.
Summary of its novel POCT ideas here:
Conventional health economic and outcomes research (HEOR) analyses of point-of-care testing (POCT) emphasize turnaround time, workflow efficiency, and operational convenience. In the present project, a paired conceptual analysis applies Shannon information theory and modern decision-theoretic frameworks to re-examine POCT as an information-processing system rather than a logistics innovation.
Using qualitative reasoning and simplified mathematical formulations, the essays model the diagnostic encounter as a communication channel linking latent disease states to clinical action. The analysis demonstrates that POCT fundamentally restructures this channel by reducing information loss (“erasures”), preserving clinical context, enabling feedback-driven sequential decision-making, and aligning diagnostic information with disease dynamics. Even when analytic sensitivity and specificity are identical to central laboratory testing, POCT increases effective information yield, improves physician–patient communication fidelity, and raises the expected utility of diagnostic information. At the system level, same-visit clarification reduces care-pathway entropy, improves triage accuracy, and stabilizes downstream resource allocation.
This new information-theoretic framing provides a unifying explanation for observed POCT advantages reported in the [conventional] HEOR and implementation literature and offers a complementary theoretical foundation for evaluating POCT value beyond speed, cost, or assay performance alone.