ct_bergstrom, (edited )
@ct_bergstrom@fediscience.org avatar

People keep telling me that is amazing for proofreading text and improving scientific writing.

I just gave a section of a grant proposal and it made 11 suggestions, none of which were worth keeping (often adding or removing a comma, or repeating a preposition in a list).

More interestedly, a number of its suggestions were identical to my originals.

JoeGrowling,
@JoeGrowling@todon.eu avatar

@ct_bergstrom everything you feed into one of these 'AI' things is likely to be regurgitated at a future date for some other user.

I don't understand how anyone would want that.

Gojira1000,

@ct_bergstrom GPT is shit - cold, runny shit

FullOnElectric,

@ct_bergstrom for all the AI hype have they distinguished the difference between and ? Automation requires significant processing capacity. Intelligence?

GregDance,
@GregDance@mastodon.green avatar

@ct_bergstrom

That’s exactly how a ToryGPT would work!

SimonDHeyes,
@SimonDHeyes@ecoevo.social avatar

@ct_bergstrom Yea I asked it to rewrite one of my blog posts and it was at the level of a 15 year old school student or your average tabloid journalist.

noyes,
@noyes@mastodon.online avatar

@ct_bergstrom

Kinda makes you wonder if using AI to catch students using AI is such a hot idea.

stefano,
@stefano@bsd.cafe avatar

@ct_bergstrom I've experienced the same situation. More, it sometimes wants me to correct something right, pushing me to make mistakes.
While I think it can be useful, people should stop using it like the oracle of delphi. It's a convincing liar.

NIH_LLAMAS,
@NIH_LLAMAS@mastodon.social avatar

@ct_bergstrom So in addition to mimicking a co-author with bad ideas, it can mimic a boss with no ideas who repackages your ideas as their own. The future is now!

twitskeptic,

@ct_bergstrom Curious what command you gave it to review the text.

ct_bergstrom,
@ct_bergstrom@fediscience.org avatar

@twitskeptic

"Please edit this for grammar:"

twitskeptic,

@ct_bergstrom I tried one of your examples using your command and it thought it was fine. So no idea, unless you submitted a very long document (context window).

Also tried grammar correction on a New Yorker article for fun and it suggested a couple of very undesirable changes. Not impressed.

ct_bergstrom,
@ct_bergstrom@fediscience.org avatar

@twitskeptic I gave it a full page of text, and I suppose that is what did it.

twitskeptic,

@ct_bergstrom FWIW, I think the smallest context window on GPT-4 is around 6000 words so it should be able to handle one page. 🤷‍♂️

PeoriaBummer,

@ct_bergstrom I find it stunning that we’re trusting our creative work to the very same organization that had no scruples about swallowing all the IP on the internet.

burritojustice,
@burritojustice@mastodon.social avatar

@ct_bergstrom ChatCTB

ct_bergstrom,
@ct_bergstrom@fediscience.org avatar

I recognize that the self-attention layer of these models allows a level of recurrence, but I wonder if this failure to recognize that it's suggestions are identical to the original is in some way a symptom of the fact that these are basically feed-forward architectures.

RWerpachowski,

@ct_bergstrom I think more than one query might be needed to get good results.

dryak,
@dryak@mstdn.science avatar

@ct_bergstrom
On the other hand if your original is good enough, no need to change it for the sake of change (See Queen's Duck in Battle Chess). Your original good enough sentence will end up in the suggestion.

BTW: have you also tried DeepL Write tool?

ook_simon,

@ct_bergstrom The replies are interesting too, they explain nothing and are only meant to generate the Eliza effect

g,

@ct_bergstrom You are ChatGPT 5.

On a serious note, people that are top of their game will of course not have a use for ChatGPT because ChatGPT is human knowledge, not the best human knowledge, just a smart person.

ewen,

@ct_bergstrom

They've set the bar pretty low for defining AI as useful. Much like the vitamins and supplements industry.

jonhendry,
@jonhendry@iosdev.space avatar

@ct_bergstrom

But have you tried writing poorly?

a32,
@a32@social.tchncs.de avatar

@ct_bergstrom chatGPT's output looks like bullshit.

lorrden,
@lorrden@c.im avatar

@ct_bergstrom you are obviously too good at writing 😊

  • All
  • Subscribed
  • Moderated
  • Favorites
  • ChatGPT
  • ngwrru68w68
  • DreamBathrooms
  • khanakhh
  • magazineikmin
  • InstantRegret
  • ethstaker
  • thenastyranch
  • Youngstown
  • rosin
  • slotface
  • osvaldo12
  • everett
  • kavyap
  • Durango
  • megavids
  • cubers
  • tester
  • GTA5RPClips
  • modclub
  • mdbf
  • cisconetworking
  • tacticalgear
  • Leos
  • normalnudes
  • anitta
  • provamag3
  • JUstTest
  • lostlight
  • All magazines