I have to say that those calculations provided by ChatGPT are remarkably like those we use (or used to) when working out how to fit a mount and camera to an aircraft or helicopter or indeed any other vehicle. Cameras are much lighter now as are the mounts. I have a mini Moy head in my loft and although it was a lightweight at the time it still weighs around 30kg and thzt is before the remotes and camera are added.
Those calculations have appeared in various documents. I recall seeing them in some UK CAA or LAA publications. So ChatGPT obviously picked them up. It isn’t actually doing any calculations; it works by clever word and phrase storage and manipulation.
Emir wrote:
. It’s clueless about any useful stuff and provided answers and advices can be equally right or wrong.
Exactly. It is a marvellous bullshitter, it talks with confidence about areas it has no real understanding of, whih is the very definition of BS.
Here is a great example of it failing spectacularly at a simple task:
It confidently states 41789 is prime (it is 11 × 29 × 131). It then states a wrong algorithm (it says “we need to check if 41789 is divisible by any prime numbers greater than 3” but should include 3) an then fails to apply it (it stops at 7).
Every ten year old should recognise this as wrong.
Even more hilarious – if you tell it that it is wrong it often tries again. Sometimes it fixes it, sometimes it makes it worse
Looks convincing, except is is using rounding so the result it utter BS
So let’s challenge it.
Doesn’t quite “get it”, does it?
Cobalt wrote:
It is a marvellous bullshitter, it talks with confidence about areas it has no real understanding of, whih is the very definition of BS.
Absolutely. A bit like a mediocre student who memorised a few basic rules and is trying to pass an exam by confidently plugging these rules everywhere, even where they don’t apply.
An older version? GPT4 writes a Python script to answer.
Yes, 3.5. If I wanted to pay for BS I would hire McKinsey ;-)
My point is not that it can’t factor numbers. My point is that it speaks with confidence but has no real understanding.
It is a fantastic tool but requires a user who knows the limitations. It will revolutionise many services, especially those which are in a well defined field of knowledge.
It will revolutionise many services, especially those which are in a well defined field of knowledge.
Or the ones that require no knowledge at all.
loco wrote:
An older version? GPT4 writes a Python script to answer.
Try with a somewhat more complicated (or uncommon) problem and the python code will look exactly like the calculations above. It will have no logic, no meaning. Also, responding with a code that simply invokes an external library, without detailing the algorithm or the calculations, is pretty laughable as far as answers go.
I remember asking it to do some meta-programming in Prolog and it just did a program returning a string which was a TIC TAC TOE program code. Thanks, bro.
maxbc wrote:
Also, responding with a code that simply invokes an external library, without detailing the algorithm or the calculations, is pretty laughable as far as answers go.
I haven’t been programming in 15 years, but I find things like shown below impressive.
Well yes but you could find the same thing on stackoverflow.com, which is indeed almost certainly where it came from. Plus, it’s not very good code – for example it tries all the even numbers as factors, which is a complete waste of time.
To me ChatGPT (etc) are just like Google, except that G just gives you a list of pages to go and look at yourself, whereas C takes all those pages, plus some other stuff it “knows”, and tries to stitch together a coherent response integrating them. Coherent, maybe, worth having – your mileage will certainly vary.
