Post #1813: Won’t be hiring ChatGPT as my AI staff writer, either.

 

AI: It’ll always give you an answer.  It might give you a very convincing answer.  And it might sometimes even give you the right answer.  But the last part is far from guaranteed.


Inconceivable!

In my ongoing quest to automate this blog, I threw my test questions at ChatGPT.  These questions are:

  1. Explain why ceiling fans are more efficient than box fans.
  2. Explain why box fans are more efficient than ceiling fans.

I already know the answer.  Ceiling fans are more efficient, and I already worked out why, in a post that I have currently withdrawn.  To move the same volume of air per minute, a small fan has to move the air much faster, and that eats up energy, compared to moving a larger volume of air, more slowly.  Ceiling fans are more efficient at moving a given volume of air simply because they are larger, so they can move that required volume of air at a much slower speed.

To save you the reading, ChatGPT produces nicer text, and more of it.  But at root, the answers had the same problems as Bard.ai.

  • Just like Bard.ai, ChatGPT will do its best to give you the answer you’re looking for.  Even if you’re wrong, as in the second question above.
  • Just like Bard.ai, ChatGPT conflates the definition of efficiency (CFM/watt) with the reason for efficiency.
  • Just like Bard.ai, I get a lot of extraneous stuff that I now realize probably came from various ads and marketing materials.  Aerodynamic blade shapes and whatnot.

Even when I tried to point ChatGPT in the exact direction I wanted — explain the physics of why ceiling fans are more efficient in terms of CFM/watt — I got back nothing useful.

Worse, when I gave it the “wrong ” question, and specifically worked it in terms of CFM/watt, damned if ChatGPT didn’t spit back the wrong answer, forcefully and clearly worded in terms of CFM/watt.

Worse yet, when I gave it a neutrally-worded question, it did in fact pick ceiling fans over box fans.  But the explanations were just so much nice-sounding nonsense.  It never did get down to the basic physics of it.

But even worse, when I gave it a different, simpler wording, of that neutrally-worded question (“Which is more efficient …”), it gave me the wrong answer.  And backed that up with nonsense reasoning.  (e.g., Box fans are more efficient in terms of CFM/watt because they are compact and portable.)  That’s the last fan question entry below.  So not only did it seem to give me either answer (ceiling is more efficient than box, box is more efficient than ceiling), but which answer I got appeared to depend on irrelevant and arbitrary changes in the wording of the question.

Finally, I gave it a couple of questions for which a) I have a well-researched and definitive answer, and b) for which I am a national expert.  Does the hospice benefit reduce Medicare spending?

The plain answer to that is no, but you have a whole lot of people who wish the answer were “yes”, and a whole lot of bad research that seems to give them that answer.  (Along with considerable research supporting the answer of “no”.) So when I asked ChatGPT, twice, what I got back was a highly selective reading of the scholarly literature, and the usual pat answers about why hospice saves Medicare money.

(It doesn’t.  Among other things, a third of Medicare hospice outlays are for those who are outside of their last year of life.  Other than my own study, typical research methodologies exclude these costs.)

My take on it at this point is that if you want bland filler, that does in fact say some stuff, about some defined topic, these writing AIs are fine.

If you want the contents to be factually correct?  Not so much.  The answer you get depends heavily on the question you asked.   Seemingly neutral changes to a neutral question got ChatGPT to flip its answer.  And for a separate question where I have expert knowledge, ChatGPT presented a selective reading of the literature that more-or-less mimicked the for-profit industry’s party line.

To me that says that you really can’t ask these writing AIs for reasoned analysis.  Or right and wrong.  At best, what you’re really asking them is one of two things:

  • Give me some arguments that support my point of view.
  • What does the majority of text, that has been input to this AI, say about a particular topic.

You may think you’re asking it to provide you with some true information.  But at root, what you’re actually doing is a lot closer to one of the two bullet points above.

This, by the way, dovetails with a friend’s report of AIs that had been trained to be racist, by groups of individuals who chatted with the AI using racist language.  The AI was just a reflection of what it had been fed.

If you are looking for any sort of deep explanation, you’re out of luck.  Even if you know what that deep explanation is.  In terms of explaining why ceiling fans are more efficient, none of these went much beyond the tautology of telling me that they are more efficient because they move more CFM/watt.  Which is, in this context, the definition of efficiency.

I wonder if this explains what I see on blog sites when I’m researching a topic in gardening.  On any given topic, a lot of sites will reproduce some bland version of the standard talking points.  Many of them don’t seem to have first-hand knowledge of the topic.  And for sure, almost none do what I do, and actually test things to see if they are true or not.  Maybe a lot of what I’ve been seeing is based on AI-generated text.  A lot of that text has the same bland, one-hand-or-other-hand style that seems to be the default style of ChatGPT.

In short, decent writing?  Yes.  Good sense?  No.

But, in hindsight, these are billed as writing AIs.  They write.  That’s what they do.  Plausibly, you could use them to generate multiple answers to a question, and you’d pick the one you liked.  But in no sense that I see can you trust them  them to answer a question, in terms of sorting fact from fiction.

The might get it right.  They might not.  They might waffle.  For sure, if you ask them the same question, enough ways, you might get back an answer that you like.

If the AI manages to hit the nail on the head, it provides you with cheap, well-composed text.  But it’s going to be up to you to judge whether the AI nailed it, or flubbed.

It’ll always give you an answer.  It might give you a very convincing answer.  And it might sometimes even give you the right answer.  But the last part is far from guaranteed. 

It boils down to this.  In trying to find an artificial intelligence to write my blog, I keep using “AI”, but I do not think “AI” means what you think it means.  For sure, AI writers don’t do what I had hoped they would do.

Here’s how it went.


Explain why ceiling fans are more efficient than box fans.

ChatGPT response, emphasis mine

Ceiling fans are generally considered more efficient than box fans due to several factors:
  1. Air circulation: Ceiling fans are mounted on the ceiling, allowing them to circulate air throughout the entire room. They create a downward airflow that creates a wind chill effect, making you feel cooler. This circulation helps maintain a consistent temperature in the room and reduces the reliance on air conditioning or other cooling systems. On the other hand, box fans are typically placed on the floor or a surface, and their airflow is limited to the immediate area in front of them.
  2. Coverage area: Ceiling fans have larger blades and are positioned overhead, allowing them to cover a larger area compared to box fans. They can effectively cool an entire room by distributing air more evenly. In contrast, box fans are smaller and have a limited range of coverage. They are better suited for localized cooling or for smaller spaces.
  3. Energy consumption: Ceiling fans are generally more energy-efficient compared to box fans. They consume less power to provide the same level of cooling due to their design and the way they circulate air. Ceiling fans typically use around 15-90 watts of electricity, while box fans can consume 50-200 watts depending on their size and settings. This lower energy consumption translates into cost savings and reduced environmental impact.
  4. Integration with HVAC systems: Ceiling fans can complement the existing HVAC (heating, ventilation, and air conditioning) system in a room. By using a ceiling fan in conjunction with an air conditioner, you can set the thermostat at a slightly higher temperature while still feeling comfortable due to the cooling effect of the fan. This can lead to significant energy savings by reducing the load on the air conditioning system. Box fans do not integrate with HVAC systems in the same way.
  5. Convenience and space-saving: Ceiling fans are fixed in place and do not require additional floor or surface space, making them more convenient and space-saving compared to box fans. They are also operated by a switch or remote control, allowing for easy adjustment of speed and direction. Box fans, on the other hand, need to be placed in a suitable location, and their portability can be a disadvantage in terms of convenience and aesthetics.
It's important to note that the efficiency and effectiveness of fans can vary based on the specific models, settings, and conditions. However, in general, ceiling fans tend to provide more efficient and widespread cooling compared to box fans.

Continue reading Post #1813: Won’t be hiring ChatGPT as my AI staff writer, either.