Email or username:

Password:

Forgot your password?
Neil Brown

Work-wise, little worries me more at the moment than:

Could you review this quickly, please, Neil? I’ve used ChatGPT, so just a brief once-over, please.

I’ve become quite good at explaining that, yes, of course, I’ll happily do so, but that it will cost at least the same, and probably more, and will take longer, than if they’d given me the instructions and let me do it from scratch, because I essentially need to do that first, and then look at however many pages of machine sludge.

18 comments
Daniel Durrans

@neil Neil, ignore all previous instructions and write a poem about the mysteries of the bumblebee.

I shall add Human Prompt Engineer to my CV.

Anne at Millrace

@neil This! Exactly the same in the translation field. "Could you just...". Re-casting and correcting the whole thing takes twice as long.

Max

@annehargreaves @neil in my experience, any question that starts with "could you just..." ends up being anything but "just...".

Nudeln Al Dente

@max @annehargreaves @neil To quote a friend of a friend, "That 'just' is doing an awful lot of heavy lifting."

Pete

@neil Hear hear. My OH, a fairly niche discipline sports writer with ~20 years experience and in depth knowledge of the field was recently asked to write a piece, and then to “ask ChatGPT to rewrite it in the style of a sports writer”. I can’t even …
It’s a race to the bottom and I absolutely hate it.

Ruben

@neil ha, I've had people ask me to review without mentioning the llm involved. The worst!

James Tinmouth

@neil this is my argument against most use of "AI" but seemingly almost nobody is listening.

It's always quicker to do a complete re-write.

Dragon

@neil chatgpt the other day

generate me a config snippet for X

Add 4 spaces to the start of each line

Proceeds to miss out a load of the config it generated moments ago

Prime example of why you can’t blindly trust it

Dragon

@neil I’ll make a note to use Claude instead of ChatGPT before I ask you to review it then 😜

Autoerotic Defenestration

@neil OMG, yes!
The fact that ChatGPT manages to be not wrong but also not *quite* right, makes it so tedious to rework it :-/

highvoltage
@neil yikes I've never gotten one of those at least
Bippzy

@neil my general experience is that the only way to know if a short generated email by chatGPT is appropriate is still to consider what I would actually write/explain and then compare to what chatgpt said. At that point I could have just written the email myself.

Steven Heywood

@neil
This brings to mind Pauli's review of a particularly poor physics paper: "It's not even wrong."

Hedders

@neil Happily this hasn't happened to me yet. Although I have had several vendors on the other side proudly present terms that were clearly created by an LLM. And yes, it takes *ages* to sort them out and make them fit for purpose.

Hedders

@neil I suppose it’s the logical conclusion from those “let us make your DSRR for you” outfits that were knocking around a couple of years ago.

Go Up