#51  
Old 02-02-2023, 06:53 PM
Ari's Avatar
Ari Ari is offline
I read some of your foolish scree, then just skimmed the rest.
 
Join Date: Jan 2005
Location: Bay Area
Gender: Male
Posts: XMDCCCLXVIII
Blog Entries: 8
Default Re: ChatGPT

Unfortunately at the moment there’s no way to prevent shorts from showing up in my subscription feed, and YouTube has been heavily pushing Shorts, so creators who want to stay at the top of the algorithm need to be making shorts, which means my Subscription feeds is becoming more often than not a wall of shorts with actual videos interspersed like gems I must wade through the shorts to find.

(This does seem like a very Google way to annoyingly push a feature they will probably drop with little notice.)
Reply With Quote
Thanks, from:
beyelzu (02-02-2023), Crumb (02-02-2023), Ensign Steve (02-02-2023), fragment (02-03-2023), JoeP (02-03-2023), lisarea (02-02-2023), viscousmemories (02-03-2023)
  #52  
Old 02-02-2023, 07:59 PM
Crumb's Avatar
Crumb Crumb is offline
Adequately Crumbulent
 
Join Date: Jan 2005
Location: Cascadia
Gender: Male
Posts: LXMMCDXXV
Blog Entries: 22
Images: 355
Default Re: ChatGPT

Oh yes, the subscription page has a lot of usability issues because they want you to use the algorithm so they don't make any improvements to it.
__________________
:joecool2: :cascadia: :ROR: :portland: :joecool2:
Reply With Quote
Thanks, from:
Ari (02-02-2023), JoeP (02-03-2023), viscousmemories (02-03-2023)
  #53  
Old 02-02-2023, 09:27 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

I was at my team stand-up this morning and my update was that I spent yesterday on a journey of discovery trying to figure out how to accomplish a specific task. A day of Googling and StackingOverflow had not gotten me there, but I had plans to meet with someone later who could probably be of help.

So I met with that person, and she was a huge help, and she gave me exactly what I needed. At the same time, one of my teammates asked ChatGPT my question. He sent it to me after, and it was the exact same solution. Down to the same code.

I literally can't even. Is this the singularity? Is it happening right now?
Attached Images
File Type: png gpt1.png (788.0 KB, 15 views)
File Type: png gpt2.png (608.8 KB, 13 views)
File Type: png gpt3.png (771.4 KB, 13 views)
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
Ari (02-02-2023), beyelzu (02-02-2023), JoeP (02-03-2023), Kamilah Hauptmann (02-02-2023), lisarea (02-02-2023), slimshady2357 (02-03-2023), Sock Puppet (02-03-2023), viscousmemories (02-03-2023)
  #54  
Old 02-02-2023, 09:37 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

I mean in the interest of transparency, I don't know if it produced compilable, runnable code, and I kind of doubt it? But which library to import, which function calls to make, that was all I needed to get unblocked. I think it's still in a place where it can be a valuable tool for me, but not quite ready to replace me. For now, anyway. Still, it's fucking uncanny.

I've seen python generated by predictive models a few years ago, and it was gibberish. A lot of the familiar buzzwords and symbols, but unpaired braces, return statements in the middle of a function, things like that. This latest is so much closer to the real thing.
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
beyelzu (02-02-2023), Crumb (02-02-2023), JoeP (02-03-2023), Kamilah Hauptmann (02-02-2023), slimshady2357 (02-03-2023), viscousmemories (02-03-2023)
  #55  
Old 02-02-2023, 09:46 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

I watched this one last night, too:


At about 13:30, he gives it a physics problem to solve in python, and it involves the Fourier transform, and I'm thinking that if it knows anything about anything, I'm going to be seeing np.fft in its code. And then it fucking did! And then at that moment in the video, the guy that's grading it says "It's scarily good" and I was like OMG, I know, right?!

I'm seriously down the rabbit hole here, you guys. :twilightzone:

As far as using it to do your homework, depending on where you are in your academic career, the goal isn't necessarily to get correct, working code, but just to generate something that was convincingly a valiant attempt. Then you turn in it and get a B or C for at least having some idea of the general concepts.
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
Ari (02-03-2023), beyelzu (02-02-2023), Crumb (02-02-2023), JoeP (02-03-2023), Kamilah Hauptmann (02-02-2023), slimshady2357 (02-03-2023), Sock Puppet (02-03-2023), viscousmemories (02-03-2023)
  #56  
Old 02-03-2023, 07:40 AM
JoeP's Avatar
JoeP JoeP is offline
Solipsist
 
Join Date: Jul 2004
Location: Kolmannessa kerroksessa
Gender: Male
Posts: XXXVMMLXIX
Images: 18
Default Re: ChatGPT

Quote:
Originally Posted by Ensign Steve View Post
I mean in the interest of transparency, I don't know if it produced compilable, runnable code, and I kind of doubt it? But which library to import, which function calls to make, that was all I needed to get unblocked. I think it's still in a place where it can be a valuable tool for me, but not quite ready to replace me. For now, anyway. Still, it's fucking uncanny.

I've seen python generated by predictive models a few years ago, and it was gibberish. A lot of the familiar buzzwords and symbols, but unpaired braces, return statements in the middle of a function, things like that. This latest is so much closer to the real thing.
What ChatGPT lacks is any attempt to check or validate or justify its output - because it hasn't been set up to do this - it's just trying to predict "what would come next" given a question.

But programming is one example of areas where it might be very useful, even profitable, to augment GPT stuff with testing and iterative improvement, possibly with reasoning. So it can say OK you gave me an English language problem, and you gave me some test cases. Here's some code that runs and passes those tests - and here's my reasoning for why this is a good solution.

This could already be possible in the labs.
__________________

:roadrun:
Free thought! Please take one!

:unitedkingdom:   :southafrica:   :unitedkingdom::finland:   :finland:
Reply With Quote
Thanks, from:
Ari (02-03-2023), Ensign Steve (02-03-2023), Kamilah Hauptmann (02-03-2023), slimshady2357 (02-03-2023), viscousmemories (02-03-2023)
  #57  
Old 02-03-2023, 12:30 PM
viscousmemories's Avatar
viscousmemories viscousmemories is offline
Admin
 
Join Date: Apr 2004
Location: Ypsilanti, Mi
Gender: Male
Posts: XXXDCCXLVII
Blog Entries: 1
Images: 9
Default Re: ChatGPT

I've had it write a couple Python scripts for me this past couple weeks to solve specific tasks that were just "using data in this file get more data from this api and add it to the file" and it worked spectacularly well. Not perfect because as it turns out writing requirements is hard, but with a few tweaks totally usable code. One issue I had was saying (effectively) "solve for this error condition" and then it did, but then when I said "okay now solve for *this* error condition", the code it produced had allowed for the second error condition but no longer had the change it made for the first one. Does any of this make sense? It's very early here.

Also, we had quite a spat yesterday because no matter how hard I tried I could not get it to spit out an accurate formula to solve something in Excel I needed to do. And when I ask how to hide certain rows or columns conditionally, I do not appreciate being given step by step instructions on how to find and delete those rows or columns, even though it ended its advice with a warning that this was a permanent hiding method.
Reply With Quote
Thanks, from:
Ari (02-03-2023), Crumb (02-03-2023), Ensign Steve (02-03-2023), JoeP (02-03-2023), Kamilah Hauptmann (02-03-2023), slimshady2357 (02-03-2023)
  #58  
Old 02-03-2023, 07:26 PM
Crumb's Avatar
Crumb Crumb is offline
Adequately Crumbulent
 
Join Date: Jan 2005
Location: Cascadia
Gender: Male
Posts: LXMMCDXXV
Blog Entries: 22
Images: 355
Default Re: ChatGPT

An unending AI generated sitcom: Twitch
__________________
:joecool2: :cascadia: :ROR: :portland: :joecool2:
Reply With Quote
Thanks, from:
Ensign Steve (02-03-2023), JoeP (02-03-2023)
  #59  
Old 02-03-2023, 07:56 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

I finally bit the bullet and signed up for my own account. "ChatGPT, do the boring parts of my fucking job for me, please"
Attached Images
File Type: png chatgpt-interview.png (118.0 KB, 21 views)
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
Ari (02-03-2023), beyelzu (02-04-2023), JoeP (02-03-2023), mickthinks (02-03-2023), viscousmemories (02-04-2023)
  #60  
Old 02-03-2023, 07:59 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

Okay, so here's one thing. I can look at those answers and know that they are good (and/or how they could be improved/fine-tuned) because I am already an expert in the domain.

If I asked it for some legal advice or medical advice (which I know we're not supposed to do), I would have no idea how to vet the answers. We're still going to need human experts on this, so that's something.
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
beyelzu (02-04-2023), ceptimus (02-04-2023), JoeP (02-03-2023), mickthinks (02-04-2023), viscousmemories (02-04-2023)
  #61  
Old 02-04-2023, 12:07 PM
viscousmemories's Avatar
viscousmemories viscousmemories is offline
Admin
 
Join Date: Apr 2004
Location: Ypsilanti, Mi
Gender: Male
Posts: XXXDCCXLVII
Blog Entries: 1
Images: 9
Default Re: ChatGPT

I'm a paying subscriber to Dave Farley's "Continuous Delivery" channel because I learn something from every single thing he writes or says, but with that caveat this was interesting:

Reply With Quote
Thanks, from:
ceptimus (02-05-2023), JoeP (02-05-2023), LarsMac (02-04-2023), mickthinks (02-04-2023)
  #62  
Old 02-04-2023, 04:06 PM
Dragar's Avatar
Dragar Dragar is offline
Now in six dimensions!
 
Join Date: Jan 2005
Location: The Cotswolds
Gender: Male
Posts: VCI
Default Re: ChatGPT

I'm very impressed with chatGPT - it's pretty amazing how much you can do with a simple 'predict next token' model of language, and I think the results of just scaling this up and up and up have surprised a lot of people.

That said...

These language models are not thinking, or reasoning, or doing any of the same things we (seem to) be doing when we talk to one another. (Good discussion on this point: https://arxiv.org/abs/2212.03551). So I'm sort of sick of people on twitter suddenly surprised when GPT makes errors that simply illuminate this aspect.

These models also seem a lot better at generating new content that reliably reproducing information. I've chatted with former colleagues who have ran projects with LLMs to summarise literature, and their main finding was that these things tend to bullshit, confidently and convincingly. In a sense, that's exactly what they're designed to do: producing the most likely 'what comes next' value, irrespective of truth or logical/mathematical/physical consistency, is almost the definition of bullshitting!

So my predictions is that this sort of technology is mostly going to be useful for processes that don't need to be correct (so ruling out a lot of things like writing code, or answering factual questions, or being an AI assistant to 'do things'), but do need to be creative and/or can have a high failure rate.

So things like online automated help services, content generation (do you write bullshit for a living? Be worried!), procedural generation in video games (or even online RPGs!), idea generation...exciting times to come in the creative industries, and that's before we talk about image and SFX models (also very impressive).
__________________
The miracle of the appropriateness of the language of mathematics for the formulation of the laws of physics is a wonderful gift which we neither understand nor deserve. -Eugene Wigner
Reply With Quote
Thanks, from:
Ari (02-04-2023), ceptimus (02-05-2023), Crumb (02-06-2023), JoeP (02-05-2023), Kamilah Hauptmann (02-04-2023), slimshady2357 (02-05-2023), Sock Puppet (02-08-2023), specious_reasons (02-04-2023), viscousmemories (02-05-2023)
  #63  
Old 02-05-2023, 01:14 PM
viscousmemories's Avatar
viscousmemories viscousmemories is offline
Admin
 
Join Date: Apr 2004
Location: Ypsilanti, Mi
Gender: Male
Posts: XXXDCCXLVII
Blog Entries: 1
Images: 9
Default Re: ChatGPT

Yeah, maybe what makes a great creative writer is the ability to produce unpredictable combinations of words in a way that still embeds meaningful expressions of human experience that humans can decipher but which AI is a long way from understanding. I can imagine AI being just as good at assembling improbable combinations of words as it is at producing predictable patterns, but I suspect its a long way off from being able to imbue unpredictable text with relatable subtext. Then again we humans are so good at spotting patterns and deriving meaning from the meaningless, maybe we would revere word salad from AI as great poetic genius.
Reply With Quote
Thanks, from:
Ari (02-05-2023), beyelzu (02-05-2023), ceptimus (02-05-2023), Dragar (02-05-2023), fragment (02-05-2023), JoeP (02-05-2023), Kamilah Hauptmann (02-05-2023), LarsMac (02-05-2023)
  #64  
Old 02-05-2023, 08:28 PM
Kamilah Hauptmann's Avatar
Kamilah Hauptmann Kamilah Hauptmann is offline
Shitpost Sommelier
 
Join Date: Mar 2016
Posts: XVMCMXLV
Default Re: ChatGPT



image.png
__________________
Peering from the top of Mount Stupid

:AB: :canada:
Reply With Quote
Thanks, from:
beyelzu (02-06-2023), ceptimus (02-06-2023), Crumb (02-06-2023), mickthinks (02-06-2023), slimshady2357 (02-06-2023), viscousmemories (02-06-2023)
  #65  
Old 02-06-2023, 06:48 AM
-FX-'s Avatar
-FX- -FX- is offline
Forum gadfly
 
Join Date: Feb 2011
Location: In your head
Gender: Male
Posts: MMLXIV
Blog Entries: 1
Default Re: ChatGPT

That is simply awful. It's not awful like a badly written piece by a struggling human, it's awful because it is just stupid, unimaginative, and meaningless.
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""

- Richard P. Feynman
Reply With Quote
  #66  
Old 02-06-2023, 06:45 PM
beyelzu's Avatar
beyelzu beyelzu is offline
simple country microbiologist hyperchicken
 
Join Date: Sep 2004
Location: georgia
Posts: XMVDCXXXVIII
Blog Entries: 1
Images: 8
Default Re: ChatGPT

On creativity:

Perhaps, I have an overly romantic idea of what constitutes creativity, but it seems to be that when AI like ChatGPT verges into seemingly creative behavior it’s slippage or mistake on the AI’s part. The better the program performs the more banal and obvious to everyone the answer should be.

A user of the program certainly could look through query answers for interesting slippages and combine them in creative ways, but that wouldn’t mean the program was creative itself.

Like I said, maybe I have an idealized view of creativity.
__________________
:blowkiss: :beloved: :blowkiss: :beloved: :blowkiss: :steve: :blowkiss: :beloved: :blowkiss: :beloved: :blowkiss:
Reply With Quote
Thanks, from:
ceptimus (02-06-2023), Ensign Steve (02-06-2023), JoeP (02-06-2023), Kamilah Hauptmann (02-06-2023), viscousmemories (02-07-2023)
  #67  
Old 02-06-2023, 07:35 PM
ceptimus's Avatar
ceptimus ceptimus is offline
puzzler
 
Join Date: Aug 2004
Location: UK
Posts: XVMMDCCCXV
Images: 28
Default Re: ChatGPT

Is it even a sonnet? A sonnet should have 14 lines with (simplistically) 10 syllables per line, and some form of rhyme scheme connecting groups of lines - often three groups of four lines, followed by a final rhyming couplet.

It does put a capital letter after (roughly) every ten syllables, so let's write it out in the usual format, but with line numbers and syllable counts:

01 Beneath the moonlit stream the beaver toils (10)
02 With gnawing jaws and sturdy webbed feet (9)
03 He builds his dam with branches and with soils (10)
04 A fortress strong to keep his kin complete (10)
05 With fur as dark as night and eyes that shine (10)
06 He works with purpose driven by his will (10)
07 To shape the land and make it truly thine (10)
08 A home for him and all his kin to fill (10)
09 But man with greed and malice in his heart (10)
10 Has sought to bring the beaver to its end (10)
11 With traps and guns he tears the dam apart (10)
12 And leaves the beaver with no home to tend (10)
13 But still the beaver builds with hope and grace (10)
14 A symbol of our Earth's enduring place (10)

I guess we're supposed to say 'webb-ed' on line two, to get the meter right :shrug:
... and the rhymes are pretty good. :yup:

I give it :star::star::star::star::star: for being an impressive example of a state-of-the-art language prediction model.

But only :star::halfstar::nostar::nostar::nostar: as a poet.
__________________
Reply With Quote
Thanks, from:
Crumb (02-07-2023), Dragar (02-08-2023), Ensign Steve (02-06-2023), JoeP (02-06-2023), Kamilah Hauptmann (02-06-2023), slimshady2357 (02-07-2023), viscousmemories (02-07-2023)
  #68  
Old 02-06-2023, 07:42 PM
Ensign Steve's Avatar
Ensign Steve Ensign Steve is offline
California Sober
 
Join Date: Jul 2004
Location: Silicon Valley
Gender: Bender
Posts: XXXMMCXLVIII
Images: 66
Default Re: ChatGPT

In the Rob Miles video, he talks about that specific misalignment. Because of how the reward function is implemented, the goal isn't to write a sonnet, the goal is to get a "this is a good answer" reinforcement from a human, who may not even know what a sonnet is. The simulation might actually "know" the rules of constructing a sonnet, but the simulacrum has "learned" that it gets rewarded by making a nice poem that isn't a sonnet.
__________________
:kiwf::smurf:
Reply With Quote
Thanks, from:
beyelzu (02-06-2023), ceptimus (02-06-2023), Crumb (02-07-2023), JoeP (02-06-2023), Kamilah Hauptmann (02-06-2023), slimshady2357 (02-07-2023), viscousmemories (02-07-2023)
  #69  
Old 02-06-2023, 07:50 PM
beyelzu's Avatar
beyelzu beyelzu is offline
simple country microbiologist hyperchicken
 
Join Date: Sep 2004
Location: georgia
Posts: XMVDCXXXVIII
Blog Entries: 1
Images: 8
Default Re: ChatGPT

We will reach the lowest common denominator with the greatest of efficiencies. You thought human executive producers were exploitive, ripped off and recycled ideas, wait til you see some Fall’s AI produced lineup :)
__________________
:blowkiss: :beloved: :blowkiss: :beloved: :blowkiss: :steve: :blowkiss: :beloved: :blowkiss: :beloved: :blowkiss:
Reply With Quote
Thanks, from:
Crumb (02-07-2023), Ensign Steve (02-06-2023), JoeP (02-06-2023), Kamilah Hauptmann (02-06-2023), Pan Narrans (02-07-2023), slimshady2357 (02-07-2023), viscousmemories (02-07-2023)
  #70  
Old 02-06-2023, 10:14 PM
LarsMac's Avatar
LarsMac LarsMac is offline
Rambling Old Fart
 
Join Date: Nov 2012
Location: On the Road again
Gender: Male
Posts: MMMCLXXX
Default Re: ChatGPT

Quote:
Originally Posted by -FX- View Post
That is simply awful. It's not awful like a badly written piece by a struggling human, it's awful because it is just stupid, unimaginative, and meaningless.
Not that it's Nobel material, but, It's not THAT bad.
__________________
“No amount of belief makes something a fact. ” ― James Randi
Reply With Quote
Thanks, from:
Crumb (02-07-2023), Kamilah Hauptmann (02-06-2023)
  #71  
Old 02-06-2023, 11:58 PM
Kamilah Hauptmann's Avatar
Kamilah Hauptmann Kamilah Hauptmann is offline
Shitpost Sommelier
 
Join Date: Mar 2016
Posts: XVMCMXLV
Default Re: ChatGPT

He’s the Mikey of the internet, he hates everything. :)
__________________
Peering from the top of Mount Stupid

:AB: :canada:
Reply With Quote
Thanks, from:
beyelzu (02-07-2023), Crumb (02-07-2023), JoeP (02-07-2023)
  #72  
Old 02-07-2023, 07:29 AM
JoeP's Avatar
JoeP JoeP is offline
Solipsist
 
Join Date: Jul 2004
Location: Kolmannessa kerroksessa
Gender: Male
Posts: XXXVMMLXIX
Images: 18
Default Re: ChatGPT

Quote:
Originally Posted by Kamilah Hauptmann View Post
He’s the Mikey of the internet, he hates everything. :)
*pretends* to hate everything, desperately chasing that high
__________________

:roadrun:
Free thought! Please take one!

:unitedkingdom:   :southafrica:   :unitedkingdom::finland:   :finland:
Reply With Quote
Thanks, from:
Crumb (02-07-2023), Kamilah Hauptmann (02-07-2023), LarsMac (02-07-2023), slimshady2357 (02-07-2023)
  #73  
Old 02-07-2023, 09:23 AM
Kamilah Hauptmann's Avatar
Kamilah Hauptmann Kamilah Hauptmann is offline
Shitpost Sommelier
 
Join Date: Mar 2016
Posts: XVMCMXLV
Default Re: ChatGPT

Next stop: Managing Ibuprofen levels to shitpost through the arthritis flareups.

Time, it comes for us all.
__________________
Peering from the top of Mount Stupid

:AB: :canada:
Reply With Quote
Thanks, from:
BrotherMan (02-07-2023), JoeP (02-07-2023), LarsMac (02-07-2023), Pan Narrans (02-07-2023)
  #74  
Old 02-07-2023, 10:43 PM
Kevlar's Avatar
Kevlar Kevlar is offline
High tech redneck
 
Join Date: Feb 2006
Location: Portland, Oregon
Gender: Male
Posts: MMMDCXLIII
Images: 16
Default Re: ChatGPT

Quote:
Originally Posted by LarsMac View Post
Quote:
Originally Posted by -FX- View Post
That is simply awful. It's not awful like a badly written piece by a struggling human, it's awful because it is just stupid, unimaginative, and meaningless.
Not that it's Nobel material, but, It's not THAT bad.
He's projecting
Reply With Quote
Thanks, from:
JoeP (02-08-2023), Kamilah Hauptmann (02-08-2023)
  #75  
Old 02-08-2023, 05:07 PM
-FX-'s Avatar
-FX- -FX- is offline
Forum gadfly
 
Join Date: Feb 2011
Location: In your head
Gender: Male
Posts: MMLXIV
Blog Entries: 1
Default Re: ChatGPT

"With traps and guns he tears the dam apart"

This is where an AI just fails,
on a level no human would
A child can grasp with ease the fault
The AI misunderstood

The subtle things we never state
We know them from our youth
The computer misses out on this
And stumbles over truths

Simple measures, a rule of thumb
Our basic humanity
Deeply known but seldom spoke
The AI can never see
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""

- Richard P. Feynman
Reply With Quote
Thanks, from:
ceptimus (02-08-2023)
Reply

  Freethought Forum > The Marketplace > Computers & Technology


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

 

All times are GMT +1. The time now is 12:08 AM.


Powered by vBulletin® Version 3.8.2
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Page generated in 0.53318 seconds with 14 queries