OpenAI Makes ChatGPT 'More Direct, Less Verbose'

Kyle Wiggers reports via TechCrunch: OpenAI announced today that premium ChatGPT users -- customers paying for ChatGPT Plus, Team or Enterprise -- can now leveraged an updated and enhanced version of GPT-4 Turbo, one of the models that powers the conversational ChatGPT experience. This new model ("gpt-4-turbo-2024-04-09") brings with it improvements in writing, math, logical reasoning and coding, OpenAI claims, as well as a more up-to-date knowledge base. It was trained on publicly available data up to December 2023, in contrast to the previous edition of GPT-4 Turbo available in ChatGPT, which had an April 2023 cut-off. "When writing with ChatGPT [with the new GPT-4 Turbo], responses will be more direct, less verbose and use more conversational language," OpenAI writes in a post on X.

Need to keep the stupidly verbose variant..

[Junta]

For the kids to keep using it to cheat on paper writing, it's got to preserve that pointlessly verbose style.

Re:

[omnichad]

I hated writing papers in school. I would write a 3 page paper and it would only take 1 page because I can't help but be concise. I would then have to painstakingly pad it out. Because there was no more to say.

Re:

[stealth_finger]

Fuck, I make these things and people freak out if there's more than like 4 lines of text.

Re:

[Junta]

To be honest, it is rare that the audience for that communication *all* have an interest in knowing the details. Meanwhile, on the consuming end of it, you have people for whom your communication is just one of hundreds they have to deal with.

I know a guy at work like this and written or spoken aloud, he spends 98% of his content describing details that don't matter and speculations of what might be happening that are way off base. He views the long text as proof of how hard he worked and thought about it,

Re:

[omnichad]

He views the long text as proof of how hard he worked and thought about it,

It takes way more hard work to summarize if there really is a lot of information involved. I write very concisely but if I have to convey a mountain of information, I will write the long thing and then spend twice as long shrinking it. If I feel the information is really necessary, it comes after the "tl;dr" summary.

Re:

[Opyros]

Pascal once wrote in a letter [quoteinvestigator.com] that he'd made it long because he lacked the time to make it short.

Re: Need to keep the stupidly verbose variant..

[ArmoredDragon]

Meanwhile in the professional world, good luck getting people to read things with more text than half a page of bullet points.

There's a very good reason for that: Time. If you don't understand what I mean, then your job doesn't involve doing much in the way of reading reports or technical manuals.

Re:

[Junta]

Yeah, I remember laboring to make one sentence turn into 500 words of pointless fluff. Was assured that was how professional communication just was supposed to be.

Then as I entered an actual professional career found out that communication was thankfully more concise, usually. People didn't have time to piss away with stupidly flowery speech.

Then over the last year I've started getting stupidly verbose "professional" emails, with people saying "it's so much easier to make my communication 'professional' wit

Re:

[blahabl]

Yeah, I remember laboring to make one sentence turn into 500 words of pointless fluff. Was assured that was how professional communication just was supposed to be.

Then as I entered an actual professional career found out that communication was thankfully more concise, usually. People didn't have time to piss away with stupidly flowery speech.

Then over the last year I've started getting stupidly verbose "professional" emails, with people saying "it's so much easier to make my communication 'professional' with ChatGPT, it's so useful".

I get the feeling that we're headed toward future where it'll be normal for emails to be written, run through ChatGPT to "professionalize" them blowing them up fivefold, sent, then by the person receiving them ran through ChatGPT to summarize them, and read.

Re:

[laxguy]

why bother reading the email when theyre just going to read it out loud in the "training"...

Re:

[nospam007]

'Yeah, I remember laboring to make one sentence turn into 500 words of pointless fluff. '

Like SF writers in the 50ies getting paid by the word.

Re:

[Opportunist]

As long as we judge kids by the amount of text they produce instead of quality, of course they will opt for as verbose as possible.

How..?

[Kokuyo]

Just like with humans, the I mostly encounter the wordiness when it tries answering a question it was forbidden from answering.

Ask it for ethnicity based crime statistics for example.

So what, will it go "Nu-uh!" now or give actual answers?

Re:

[Seven Spirals]

Yes, it will give you an more concise refusal. It now can summarize the government and corporate propaganda it produces. It will now more briefly imply you are racist anti-vaxxer that's guilty of wrongthink.

Re:

[fibonacci8]

There's no way it's even that close to be correct.

Re:

[ToasterMonkey]

Just like with humans, the I mostly encounter the wordiness when it tries answering a question it was forbidden from answering.

Ask it for ethnicity based crime statistics for example.

So what, will it go "Nu-uh!" now or give actual answers?

I don't see that at all with ChatGPT 4, it searched the internet for my query because it doesn't have those stats built-in, then it summarized exactly the way I asked, by race and gender. It wasn't any wordier than it needed to be, and your reasoning is flawed, should the answer to "is an electron a photon?" be a short one or a long one? With humans, you get short answers when people are tired of your dumb questions, or they can't answer. How do you not know this?

From the first row, I really need to watch o

Re:

[NobleNobbler]

This isn't really my hill to die on or anything, but if you mean because the number is bigger, then someone taking... a racially influenced view of the numbers, for lack of a better phrase, would counter with the demographics of the US.

Personally, I think it's poverty related, but who asked me

Oh boy...

[OfMiceAndMenus]

Kyle Wiggers

Oh man, high school must have been brutal for this guy.

Re:

[Seven Spirals]

Yeah, unless you're going into rap (in which case your name would be found utterly hilarious), you probably wanna change that last-name, Kyle.

It should save us a lot of time

[Opportunist]

At least it won't take that long anymore to notice the answer is total bullshit.

Re:

[gweihir]

Only for people with actual intelligence. These seem to be in shorter and shorter supply in the human race. Many of them have given up on Artificial Incompetence though.

Re:

[Opportunist]

People with sub-par intelligence already find their sages and nuggets of wisdom in self-help books.

Efficiency

[fibonacci8]

Now you can get the wrong answers to questions in fewer words.

Good news.

[MindPrison]

They're actually listening.

I've been harping on them for a year about this.

Let's test it out

[Ambigwitty]

It definitely struggles with implementing libssh2 in swift. So far, suggestions are reasonable enough, but it failed to suggest implementing a check for st_size when getting remote file stats for scp. The scp c example it is referencing from clearly does that. I can coax it into a correct implementation, but until stuff like that improves, noob mistakes will be cropping up all over the place.

Re: Let's test it out

[Ambigwitty]

Coaxing worked.. I guess next thing I want to tinker with are bleeding edge topics (e.g. code documented up to the model cutoff date).

Re:

[gweihir]

So when you already know how to do it right, you can manipulate it into eventually getting it right as well. On something this critical, that strikes me as worse than useless.

Re:

[Ambigwitty]

Depends on what the difference would be without it. If it can get there with guidance while saving significant time under close supervision, I'll give it a try. I don't think the most useful aspect here is with accuracy or safety, but it's extremely useful when starting daunting tasks, things you might procrastinate with otherwise, etc.

Re:

[gweihir]

Hmm. Does not really convince me, but well. For non-critical stuff (which anything using libssh is _not_), this may have some validity.

Chat-splaining

[VeryFluffyBunny]

Well, they've solved the Chat-splaining problem. How about having a go at the man-splaining one?

I had that nailed months ago.

[Petersko]

Just keep prompting "TL;DR" until you're happy.

ChatGPT ex plus alpha

[Z80a]

Personally, i will wait for Super ChatGPT turbo HD Remix, or maybe ChatGPT III: 3rd strike

/o\

[easyTree]

I've yet to find an AI which can answer this in the way it's intended:

"
Make me a list of domain name candidates where the whole name may be read as a single word or phrase related to the topic "time tracker app"

For example:
time.ly is the word timely assuming there's a TLD .ly

Please ensure that only valid TLDs are used
"

Actual responses to the above prompt:

-1-
Certainly! Here's a list of domain name candidates related to the topic "time tracker app", where the whole name may be read as a single word or phrase: