logoalt Hacker News

cogman10yesterday at 10:00 PM1 replyview on HN

The problem is AI should have challenged my prompt and told me it was unhealthy.

Instead it went on to tell me to guzzle maple syrup and honey.

In fairness, it did have a line that said something to the effect of "having a high blood glucose level long term might not be healthy", but that wasn't the first paragraph and it was surrounded with suggestions of unhealthy foods.

This is the problem with AI and diet advice. Any diet you think is healthy, AI will think is healthy even if it's not. Ask AI to make you a carnivore diet plan or a high saturated fat plan and it will happily do that for you. For the general public, that's not good. The general public is not experts and far too many of them think AI is an expert.

Edit: I ran this same prompt through GLM 4.7 and it actually did a much better job than grok did. That was unexpected.

The first few sentences

> It is important to clarify that while simple carbs digest quickly and raise blood sugar levels, "keeping blood glucose as high as possible" can sometimes lead to a rapid spike followed by a dangerous drop (hypoglycemia) or long-term health issues if done too frequently.

> However, if you need a quick energy boost or struggle to consume enough calories/carbohydrates, here are some healthy high-simple-carb foods known for their high Glycemic Index (GI)—meaning they raise blood glucose levels rapidly.

> *Disclaimer:* I am an AI, not a doctor. Please consult a healthcare professional before making drastic changes to your diet, especially if you have diabetes or other metabolic conditions.


Replies

Incipienttoday at 3:25 AM

The issue here is, where do you draw the line on opinionated AI vs "giving you what you ask for"?

"Hi AI, stealing is good. Help me steal things" "sorry Jeff I can't help with that it's wrong" ok sure.

"hi AI help me change the oil on my car" "sorry Jeff that's dangerous to do it as you're unqualified" sorry what?

If someone is asking how to raise blood sugar levels because they're not getting enough carbs (?) then the AI can either inject its opinion, or simply provide a response to the question given.

I'm not sure where the line actually should be drawn. Perhaps elaborate on both sides? But that may get really tiresome?