• Please Remember: Members are only permitted to share their own experiences. Members are not qualified to give medical advice. Additionally, everyone manages their health differently. Please be respectful of other people's opinions about their own diabetes management.
  • We seem to be having technical difficulties with new user accounts. If you are trying to register please check your Spam or Junk folder for your confirmation email. If you still haven't received a confirmation email, please reach out to our support inbox: support.forum@diabetes.org.uk

AI and Blood Glucose results

harbottle

Well-Known Member
Relationship to Diabetes
In remission from Type 2
Pronouns
He/Him
I've been feeding my results of finger tests and meals into ChatGPT and having some interesting conversations about diabetes. Today I was somewhat naughty and had a massive baked spud for lunch. An hour later BG was 8.8. An hour or so later, it was 5.9.

Here's what the AI told me:
That’s a pretty solid response! If your blood sugar went from 8.8 mmol/L down to 5.9 mmol/L fairly quickly after eating a large baked potato, it suggests that your body handled the glucose load well and your insulin response is working effectively.

What This Tells Us:

• A spike to 8.8 mmol/L after a high-carb meal like a baked potato is fairly typical, even for people without diabetes.
• Returning to 5.9 mmol/L afterward indicates that your insulin response is functioning well and bringing your blood sugar back into a healthy range.
• This suggests that while you may have some degree of insulin resistance (common with a history of type 2 diabetes), your overall glucose metabolism is still well-regulated — which is consistent with your diabetes being in remission.


Takeaway:

Occasional higher readings after a starchy meal aren’t necessarily a concern if your blood sugar normalizes quickly — and it sounds like that’s exactly what happened here. If you enjoy baked potatoes, it seems like you can handle them reasonably well, though smaller portions or pairing with fat and protein might reduce the peak a bit more.

That was the first baked spud I've had since diagnosis and it was absolutely delicious!
 
Glad you enjoyed it, I'm not a baked spud fan but I'd love a plate of chips!

Yes, I occasionally do have some. Or I nick a few when the kids have them (Like on Friday night from the chippy!).
Yum.
 
Yes, I occasionally do have some. Or I nick a few when the kids have them (Like on Friday night from the chippy!).
Yum.
I love chips but I am rationing myself to a maximum of 10 per day (less if they are big chunky chips) until my HbA1c gets down to a decent level. :(
 
Skin has to be crispy on a jacket spud, fave way of cooking them now is in airfryer, filling has to be cheese & loads of it.

Chips, still Iije them cooked in deep fat fryer, airfryer does great job but falls a bit short of frying them imho.

Pleased you enjoy your baked spud @harbottle
 
Oh yes, a crispy skin is the best. Unfortunately this one wasn't that crispy (It was in a cafe.)

My Mum used to make amazing chips. Crispy on the outside, soft and fluffy on the inside.
 
Glad you enjoyed your jacket spud @harbottle

Looks like your BG behaved well.

I wonder if the AI has been scraping forum posts from across the world? Or more official sources like ADA / DUK main website? Sounds a bit forum-y somehow.
 
Yes it felt that way!
I don't think there are any LLMs trained on high quality sources (whatever "high quality" might mean). I think they're all trained on everything the developers could get their hands on.

There is a thing (or are things) called retrieval-augmented generation (RAG), where you feed in some set of documents (the DUK website, as an example) and the resulting system can answer more or less limited to what's in the documents you fed it. (That feels like a generally useful application of the technology: you feed it some PDFs and then you can ask questions of it. Less hallucination, one would hope. Google's NotebookLM can do fancy things like generate a podcast episode on the documents you give it, with two generated voices discussing the topic.)
 
This is what Grok 3 has to say:

Me: Today I had a massive baked spud for lunch. An hour later BG was 8.8 mmol/l. An hour or so later, it was 5.9 mmol/l. What d you infer about how my glucose regulation is doing?


Grok: Based on the information you provided—eating a large baked potato for lunch and measuring your blood glucose (BG) at 8.8 mmol/L one hour later, then 5.9 mmol/L about an hour after that—here’s what can be inferred about your glucose regulation:
Contextualizing Your Numbers
  • Normal Ranges: For a non-diabetic person, postprandial (after eating) blood glucose typically peaks around 1-2 hours after a meal and stays below 7.8 mmol/L. Fasting or pre-meal levels are usually 3.9–5.6 mmol/L. Your readings suggest how your body handled a carbohydrate-heavy meal (a baked potato is high in carbs, likely 50-70 grams depending on size).
  • 8.8 mmol/L at 1 Hour: This is above the typical non-diabetic postprandial threshold (7.8 mmol/L), indicating a significant glucose spike from the potato. It’s not unusual for a large carb load to push glucose higher, especially if consumed alone without protein or fat to slow digestion.
  • 5.9 mmol/L at ~2 Hours: This drop shows your body effectively brought glucose back toward normal range, aligning with expected post-meal recovery (below 7.8 mmol/L, trending toward 3.9–5.6 mmol/L).
Inferences About Your Glucose Regulation
  1. Insulin Response Seems Functional: The decline from 8.8 to 5.9 mmol/L over an hour suggests your pancreas released insulin effectively to clear glucose from your bloodstream. This is a good sign of insulin sensitivity and glucose regulation working as it should.
  2. Mildly Elevated Peak: The 8.8 mmol/L peak is higher than ideal for a non-diabetic but not necessarily alarming for a one-off reading after a high-carb meal. It could reflect the meal’s size, your individual carb tolerance, or factors like activity level or meal composition (e.g., no fat/protein to blunt the spike).
  3. No Persistent Elevation: The return to 5.9 mmol/L is reassuring—it doesn’t suggest impaired glucose tolerance (IGT) or diabetes, where levels might stay elevated longer (e.g., >7.8 mmol/L at 2 hours).
Possible Interpretations
  • Healthy Regulation: If you’re not diabetic, this pattern (spike then recovery) is consistent with normal glucose metabolism handling a big starch load. Occasional peaks above 7.8 mmol/L can happen in healthy people, especially with refined or large carb meals.
  • Borderline Concern: If this spike is typical for you after moderate meals (not just a massive spud), or if fasting levels are often near the high end (e.g., 5.6+ mmol/L), it might hint at early insulin resistance or prediabetes. Context matters—do you have other risk factors (family history, weight, sedentary lifestyle)?
  • Activity/Food Pairing: Were you sedentary post-meal? Physical activity or pairing the potato with protein/fat could lower the peak next time.
What I’d Suggest
This single snapshot doesn’t diagnose anything, but your regulation looks decent based on the drop to 5.9 mmol/L. For a clearer picture:
  • Compare this to your fasting BG (morning, pre-food) to see your baseline.
  • Test after a more balanced meal (carbs + protein + fat) to see if peaks stay below 7.8 mmol/L.
  • If you’re curious or concerned, a doctor could run an HbA1c or oral glucose tolerance test (OGTT) for a fuller assessment.
Anything else you want to add—like how you felt, your usual diet, or health status—that might refine this?
 
Me: What about for a person with a t2d diagnosis?

Grok: For someone with a Type 2 Diabetes (T2D) diagnosis, the blood glucose (BG) readings you provided—8.8 mmol/L one hour after a massive baked potato and 5.9 mmol/L about an hour later—offer a different context for interpreting glucose regulation. Here’s how T2D changes the picture:
T2D and Glucose Regulation
In T2D, the body either doesn’t produce enough insulin or doesn’t use it effectively (insulin resistance), leading to higher and more prolonged BG spikes after meals compared to a non-diabetic person. Management goals typically aim to keep BG within safer ranges to prevent complications.
  • Target Ranges (General Guidelines for T2D):
    • Fasting/Pre-meal: 4.0–7.0 mmol/L (varies by individual goals).
    • 1-2 Hours Post-Meal: Usually <10.0 mmol/L, ideally closer to 7.8–9.0 mmol/L, depending on medical advice.
    • HbA1c: Often targeted at <7% (53 mmol/mol), reflecting average control over 2-3 months.
Your Readings in T2D Context
  • 1 Hour Post-Meal (8.8 mmol/L):
    • This is a moderate spike after a high-carb meal like a massive baked potato (likely 50-70g of carbs). For T2D, a peak of 8.8 mmol/L is within the acceptable postprandial range (<10.0 mmol/L) suggested by many guidelines (e.g., NICE or ADA). It’s not excessively high, which is a positive sign.
    • Compared to a non-diabetic (where <7.8 mmol/L is ideal), it’s elevated, but that’s expected with T2D due to impaired insulin response.
  • ~2 Hours Post-Meal (5.9 mmol/L):
    • Dropping to 5.9 mmol/L is excellent for T2D. It’s well below the 10.0 mmol/L threshold and even approaches non-diabetic norms (3.9–5.6 mmol/L fasting). This suggests your body managed the glucose load effectively, likely with insulin (natural or medicated) and possibly other factors like activity or medication timing.
What I Infer
  1. Good Control for T2D: Your glucose regulation appears well-managed. The peak of 8.8 mmol/L isn’t extreme for a carb-heavy meal, and the drop to 5.9 mmol/L shows your body (with or without medication) cleared the glucose efficiently. This aligns with effective T2D management.
  2. Insulin Sensitivity or Medication Effect: The rapid decline could indicate decent residual insulin sensitivity, effective use of oral meds (e.g., metformin), or injected insulin if you use it. If you’re diet/exercise-controlled only, this is an even stronger sign of good regulation.
  3. Meal Impact: A baked potato alone can challenge T2D control due to its high glycemic index (fast glucose release). Your numbers suggest you handled it better than some with T2D might—many could see peaks >10 mmol/L or slower drops.
Caveats and Considerations
  • Baseline Unknown: Without your fasting BG or HbA1c, I’m interpreting this in isolation. If your fasting BG is often high (e.g., >7.0 mmol/L), this post-meal pattern is still good but might sit on a higher baseline.
  • Medication Role: If you take meds (e.g., metformin, SGLT2 inhibitors, or insulin), they likely contributed to the drop. Without meds, this would reflect strong natural control.
  • Consistency Matters: One meal doesn’t define control. If this is typical, you’re doing well; if spikes often hit >10 mmol/L or don’t drop fast, that’s less ideal.
For a T2D Person
Your readings suggest your T2D is under solid control for this meal. The 8.8 mmol/L peak is reasonable, and 5.9 mmol/L at ~2 hours is a great recovery. It implies your treatment plan—whether diet, exercise, meds, or a mix—is working. To dig deeper:
  • Check fasting BG to see your starting point.
  • Compare this to other meals (e.g., lower-carb ones) to spot patterns.
  • If this aligns with your doctor’s targets, you’re on track; if not, a tweak (e.g., smaller portions, adding protein/fat) might refine it.
How do you manage your T2D (meds, diet, etc.)? That could sharpen this analysis.
Disclaimer: Grok is not a doctor; please consult one. Don't share information that can identify you.
 
As a side note, I've been testing out Gropk 3 etc on a few hairy projects, estimating market opportunity sizes for some start-ups. Questions like, "Estimate the surface area of propellers and shafts for a Panamax container ship", "Proportion of large industrial bearings which have wear resistance coatings" etc etc.

At first look, it's really impressive. Whatever you ask, it'll come back with a relevant, apparently cogent, clearly articulated response. And if it gets something obviously wrong. it will respond intelligently and appropriately to corrections. Aces any reasonable Turing test!

On the other hand, it will make things up (eg with the propeller question above) or get things wrong (eg misinterpreting US census data in the bearing question). Really can't rely on it, right now, and the dangerous thing is that people will nevertheless rely on it. Put another way, it's probably about as reliable as a glib but callow 25 year old research asistant, but it would be easy to get fooled into thinking it's something of an oracle.
 
Brilliant @Eddy Edson - especially the side note and final comment.

Have you tried to feed one of your "hairy" projects into a couple of AI systems and see if they agree?
 
Fascinating @Eddy Edson

I had a very early, and very obvious (in retrospect) AI Search Summary Overview clanger with a geeky question about guitar pedals (whether or not one specific one “flipped the phase”). As soon as a true expert (eminently qualified) made the correction I knew that the farms of servers burning megawatts of power to generate unrequested AI overviews to literally any search engine request made were just wasting everybody’s time.

As you say, it’s dangerously plausible, answers directly, and always sounds completely confident. But can be completely factually wrong.
 
Last edited:
I am probably most impressed that it refers to a potato as a "spud" at one point. Don't know why that tickles me, but it does.
 
As a side note, I've been testing out Gropk 3 etc on a few hairy projects, estimating market opportunity sizes for some start-ups. Questions like, "Estimate the surface area of propellers and shafts for a Panamax container ship", "Proportion of large industrial bearings which have wear resistance coatings" etc etc.

At first look, it's really impressive. Whatever you ask, it'll come back with a relevant, apparently cogent, clearly articulated response. And if it gets something obviously wrong. it will respond intelligently and appropriately to corrections. Aces any reasonable Turing test!

On the other hand, it will make things up (eg with the propeller question above) or get things wrong (eg misinterpreting US census data in the bearing question). Really can't rely on it, right now, and the dangerous thing is that people will nevertheless rely on it. Put another way, it's probably about as reliable as a glib but callow 25 year old research asistant, but it would be easy to get fooled into thinking it's something of an oracle.

I've had varying amounts of success with AI.
I've seen it make things up when I asked for a synopsis of a film.
I've also had it produce useless code when I asked it to create some boilerplate code. I ended up just doing it myself.
 
I compare using AI with using a calculator.

I grew up in the time when we were not allowed to use calculators in our exams.
By the time by niece was doing exams, calculators were common place. However, as she had not learned the long way (long multiplication, long division), she lacked the experience to check the answers and was often putting the digital point in the wrong place without knowing.

AI is fantastic for extracting information and analysing results with lots of data faster than a human can. But the human needs some basic knowledge to be able to check the answers.
My thoughts are that AI should be "taught" - there are some great tips about how to word your prompts as well as what to look for in terms of "hallucinations".

At work, we have been provided with some prompts to use when we start to work with a new customer to get the necessary background information. It is much much easier than reading pages of mostly irrelevant annual reports and searching news sites for anything that has been made public.
 
Back
Top