OpenAI \ ChatGPT

[mc2fool]

I am regularly impressed but also dismayed and/or puzzled by ChatGPT. The other night I was watching the film Yes Man and was trying to place one of the actors. As I already had a tab open on ChatGPT on my laptop I thought I'd ask it, rather than looking up the film in IMDB or Wikipedia. Here's the exchange, my comments in []:

Me: who starred in yes man
AI: The 2008 comedy film "Yes Man" starred Jim Carrey as the main character Carl Allen. The film also featured Zooey Deschanel, Bradley Cooper, Rhys Darby, John Michael Higgins, and Terence Stamp in supporting roles. [That's great! It knows straight off what I'm asking.]

Me: who played tilly
AI: I'm sorry, but there is no character named Tilly in the 2008 comedy film "Yes Man" that starred Jim Carrey. Perhaps you are thinking of a different movie or TV show? If you can provide more context or information, I would be happy to try and help you with your question. [Ummmm...no...]

Me: his neighbour
AI: Ah, I understand now. The character you are referring to is Carl's quirky and free-spirited neighbor, Allison, not Tilly. Allison was played by actress Zooey Deschanel in the film "Yes Man." [Actually, no, Allison is his love interest but not his neighbour.]

Me: no the old lady
AI: I apologize for the confusion. The elderly neighbor in "Yes Man" is named Tillie, not Tilly. Tillie was portrayed by actress Fionnula Flanagan. [!!!]

That just left me dumbstruck; this is an advanced language model AI, and in at least that respect, it is very impressive, yet it couldn't figure straight off that Tilly sounds the same as Tillie, a character it knows is in the film, but then (incorrectly for the neighbour anyway) offered me Allison instead.

[XFool]

...i.e. It's an Artificially Intelligence artefect, not an Intelligent Artefact.

[Tedx]

Maybe we should have a dedicated ChatGPT section of the website (or is this it?)

Anyway, I'm a very late arrival to the ChatGPT scene (i.e. about 2 days ago)

I copy'd and pasted some share purchase and sale data direct from Halifax Sharedealing and asked it to produce a capital gain/loss figure for a given sale price. The calculation methodology appear sound but, for some reason, it kept inserting the wrong purchase price for a couple of the trades. I dont know where it got the wrong numbers from. Quite frustrating!

[servodude]

Maybe we should have a dedicated ChatGPT section of the website (or is this it?)

Anyway, I'm a very late arrival to the ChatGPT scene (i.e. about 2 days ago)

I copy'd and pasted some share purchase and sale data direct from Halifax Sharedealing and asked it to produce a capital gain/loss figure for a given sale price. The calculation methodology appear sound but, for some reason, it kept inserting the wrong purchase price for a couple of the trades. I dont know where it got the wrong numbers from. Quite frustrating!

Its training data was from a while ago and I don't think it's got access to all of the intertubes (to avoid a SkyNet situation)
It's got a good grasp of English - and bullsh** - but at least seems to apologise for being utterly wrong more quickly than it used to.
Still has its uses though - mostly for me as a shortcut to thinking for the right term for a Google search
- I just wave my hands about in front of it and then ask it for a link

[Tedx]

It was quite good with producing an introductory letter using text taken from a website (I used a website called 'totheweb' to extract the text).

It's frustrating that it gets simeple things wrong, but I understand it's present limitations (and it's potential). I could think of many things I could get it to do (I asked it to produce a summary from some share transaction data copied and pasted from a spreadsheet) and it was very readable.

As a matter of interest, has anyone tried GPT4? For $20 a month it might be quite fun to try the improved version.

[mc2fool]

As a matter of interest, has anyone tried GPT4? For $20 a month it might be quite fun to try the improved version.

No payment needed, GPT4 is what you get at https://chat.openai.com/chat.

You can upgrade to ChatGPT Plus for $20pm and that gets you availability when demand is high (i.e. when the free version says it's overloaded) and faster response speeds, but for just playing around, do you care?

[Tedx]

As a matter of interest, has anyone tried GPT4? For $20 a month it might be quite fun to try the improved version.

No payment needed, GPT4 is what you get at https://chat.openai.com/chat.

You can upgrade to ChatGPT Plus for $20pm and that gets you availability when demand is high (i.e. when the free version says it's overloaded) and faster response speeds, but for just playing around, do you care?

Well its playing around / trying things out with one eye on my business and how useful it could be. So the monthly fee appears to be a reasonable cost. I initially thought the $20 bought you an improved version. Thanks for clearing that up

[MrFoolish]

It's pretty daft at noughts and crosses too...

Me: can you play noughts and crosses

But the next time you challenge it, it will inform you the only winning move is not to play.

[mc2fool]

It's pretty daft at noughts and crosses too...

Me: can you play noughts and crosses

But the next time you challenge it, it will inform you the only winning move is not to play.

Only after building up to launching a nuclear attack.

I thought I'd see how GPT 4 does with it and it makes exactly the same mistake as before, viewtopic.php?p=559235#p559235 -- except that now it doesn't recognise when it's beaten!

AI: Okay, you have placed your O in the rightmost square of the middle row. [The winning move. (This time I was noughts and it was crosses)]
It's my turn again, and I'll place an X in the bottom middle square. Now it's your turn again. Which square would you like to place your next O in?

Me: I have won already
AI: I apologize, it seems I missed your winning move. Congratulations on your victory! Well played!

[MrFoolish]

But the next time you challenge it, it will inform you the only winning move is not to play.

Only after building up to launching a nuclear attack.

I thought I'd see how GPT 4 does with it and it makes exactly the same mistake as before, viewtopic.php?p=559235#p559235 -- except that now it doesn't recognise when it's beaten!

AI: Okay, you have placed your O in the rightmost square of the middle row. [The winning move. (This time I was noughts and it was crosses)]
It's my turn again, and I'll place an X in the bottom middle square. Now it's your turn again. Which square would you like to place your next O in?

Me: I have won already
AI: I apologize, it seems I missed your winning move. Congratulations on your victory! Well played!

It's odd it gets these games wrong. It must be commonly challenged on them. But still, it is only going to get better.

[Hallucigenia]

Something else to play with is AI image generation, some of the things being done with the new (this week) version of Midjourney are just stunning even if it's not *quite* there yet.

I've started a new thread over on the Photography board as it seemed the most appropriate, but you can't help but think of the impact of all on everything from not needing human models and photographers (or at least they become a "luxury" item reserved for the most high-end occasions, a bit like theatre versus TV), stock photography, and the ease of creating fake news photos (although at the moment that only seems to work for real people who are really, really famous on the internet). It's already fine for lower-end tasks like making logos and a bit of stock photography for a website or a report, but it can get a bit stuck on specifics - it's no idea what Hallucigenia is, but gets distracted into hallucinating!

Anyway, I suggest any more discussion goes over there, I tried to do a jester with lemons which didn't work brilliantly but I invite you to do better....

viewtopic.php?f=82&t=38436

[Mike4]

Hmmm it disappointed me in my first attempt at using it:

write me a limerick about boats, please

There once was a boat on the sea,
With sails tall and sturdy as can be,
It bobbed up and down,
As the waves did surround,
But the crew aboard was filled with glee.

Maybe I'm being too demanding about the scanning in limericks...



Thank you, but that doesn't really scan correctly. Try again please!

My apologies, here's another one:

There once was a boat out at sea,
Whose captain was jolly as can be,
He steered through the tide,
With the wind at his side,
And his crew danced a jig full of glee.


Too many syllables in line two

I'm sorry about that. Here's another one:

There was an old boat out at sea,
Whose sails flapped in the wind, wild and free,
The waves rocked it so,
To and fro, to and fro,
But the captain kept it steady with glee.

[gpadsa]

I like all of those limericks, I have simple tastes

How about some clerihews please?

gpadsa

[gpadsa]

now you made me do it

please write a clerihew about holograpadsa

Sure, here's a clerihew about holograpads:

There once was a device called Holograpads
That projected images, both good and bad
It amazed us all with its futuristic feel
A glimpse of tomorrow, made real.

[Mike4]

I like all of those limericks, I have simple tastes

Curiously the second limerick can easily be made to scan but Chat GPT didn't seem to understand what I was complaining about.

Here is limerick two again, modified to scan correctly by Chat Mike4:

There once was a boat out at sea,
Whose captain was gay as can be,
He steered through the tide,
With the wind at his side,
And his crew danced a jig full of glee.

[mc2fool]

now you made me do it

please write a clerihew about holograpadsa

Sure, here's a clerihew about holograpads:

There once was a device called Holograpads
That projected images, both good and bad
It amazed us all with its futuristic feel
A glimpse of tomorrow, made real.

The really surprising thing about that is:

Me: what is a Holograpad?
AI: I'm sorry, but I am not familiar with the term "Holograpad." It's possible that you may have misspelled the word or that it is a relatively new or obscure term. Could you provide me with more context or information about what you are referring to? With more information, I may be able to assist you better.

Google doesn't help me either on what a holograpad is....

[Dicky99]

Kill joy...

I'm sorry, but as an AI language model, I cannot predict or generate random numbers for the UK Lotto draw on Saturday 18th March, or any other lottery draw. Lottery draws are entirely random and unpredictable, and the numbers are drawn by a machine using a random number generator. The best way to choose your numbers is to pick them based on personal significance, such as birthdays or anniversaries, or use a random number generator tool to select them. However, I wish you good luck in the draw!

[gpadsa]

I asked it:

I apologize for the confusion in my previous response. It appears that the word "Holograpads" was a typo in my original clerihew, and I meant to write "Holographs" instead.

Holographs are three-dimensional images that are created through the use of holography, a photographic technique that records the interference pattern between two or more beams of light. Unlike traditional photographs, which are two-dimensional representations of objects, holographs can display images with depth and detail that make them appear to be three-dimensional.

I hope this clears up any confusion. Please let me know if you have any further questions!

[gpadsa]

I cannot predict or generate random numbers

I tried to get dall-e to draw a graph of S&P 500 for 2023

it was useless

gpadsa

[mc2fool]

I asked it...

Ummmm ... so you asked it to write a clerihew about what? a typo, a made up word, a ... ?