I, ChatGPT
-
@DogsB said in I, ChatGPT:
@Applied-Mediocrity said in I, ChatGPT:
@DogsB I know! It's a pun. No, the other thing... palindrome.
Writing it backwards was the original idea but I thought I would get fancy and then chatgpt just sucked the will out of me so you won't get to read the joke now. Swings and roundabouts I suppose.
ChatGPT is stupid. Its keeGerawdraH
(keeGreidraH is a Klingon insult.)
-
@boomzilla said in I, ChatGPT:
I banned OpenAI because it's retarded and can't follow the rate request.
Let me help you getting rid of chinese bots too
-
-
Can't have it both ways, folks.
-
@Arantor said in I, ChatGPT:
Can't have it both ways, folks.
Rules for thee has been the modus operandi of a lot of society for quite a while.
-
-
@Arantor that’s not copyright, that’s (presumably) a trademark.
The Wikipedia page claims (also presumably without having consulted any lawyers) that “lol, this shit is too trivial for copyright, but that it’s protected by trademarks.
-
@topspin said in I, ChatGPT:
@Arantor that’s not copyright, that’s (presumably) a trademark.
The complaint was probably filed by their AI, which hallucinated that copyright and trademark are the same thing.
-
@topspin said in I, ChatGPT:
lol, this shit is too trivial for copyright, but that it’s protected by trademarks
This.
-
@Gern_Blaanston said in I, ChatGPT:
@topspin said in I, ChatGPT:
@Arantor that’s not copyright, that’s (presumably) a trademark.
The complaint was probably filed by their AI, which hallucinated that copyright and trademark are the same thing.
mistake is probably from reddit's message
-
@Bulb said in I, ChatGPT:
@kazitor said in I, ChatGPT:
If your model is still kinda shit after ingesting a million times more words than anyone could ever read in their lifetime, maybe your model actually is kinda shit.
Wasn't overtraining actually a thing? I think there was an effect when beyond certain point, more training was making the models worse.
Yes. And it was not about quality of input. Just imagine a measurement - you have say 10 pairs of x and y. In reality, the relation is linear, y=ax+b. But measurement have some deviations of course. And now you do overlearning: fit the 10 data points into y=ax^10+bx^9+c*x^8+....
-
@boomzilla said in I, ChatGPT:
dickholes
WDTWTF is a great place to learn odd english words I would never have thought of...
Is a dickhole the hole where dickshit get excreted?
-
@BernieTheBernie said in I, ChatGPT:
@boomzilla said in I, ChatGPT:
dickholes
WDTWTF is a great place to learn odd english words I would never have thought of...
Is a dickhole the hole where dickshit get excreted?A dickhole is certainly great at pissing everything off.
-
@topspin said in I, ChatGPT:
@Arantor that’s not copyright, that’s (presumably) a trademark.
The Wikipedia page claims (also presumably without having consulted any lawyers) that “lol, this shit is too trivial for copyright, but that it’s protected by trademarks.
All the subreddit people need to do, presumably, then is to ask any of the image prompting AIs to draw the logo and use whatever it comes back with, ideally if it generates something almost but not quite exactly 100%.
I was aware of the difference, but OpenAI is one of the bodies that wants their cake and eat it, either copyright doesn’t matter, as in the case of training their material; or it does, as in the matter of their logo.
-
@BernieTheBernie said in I, ChatGPT:
@Bulb said in I, ChatGPT:
@kazitor said in I, ChatGPT:
If your model is still kinda shit after ingesting a million times more words than anyone could ever read in their lifetime, maybe your model actually is kinda shit.
Wasn't overtraining actually a thing? I think there was an effect when beyond certain point, more training was making the models worse.
Yes. And it was not about quality of input. Just imagine a measurement - you have say 10 pairs of x and y. In reality, the relation is linear, y=ax+b. But measurement have some deviations of course. And now you do overlearning: fit the 10 data points into y=ax^10+bx^9+c*x^8+....
Real data is noisy. It's basically a truism that anyone in any scientific or technical field should know: reality always fucks with things in little ways all the time. Most noise follows a spectrum that is dominated by much higher frequencies than signals that you're looking for. If you overfit your predictions, you are just matching against the noise, and that's not worthwhile ever.
AIs do much higher order curve fitting when learning, and that makes matching the noise much easier! Adding more training data mostly just adds more noise. I think it is amazing that they do as well as they do, but they still can't help generating BS "at random". This is why it is vital (if you're doing custom training) to split your data into a training set and a test set (no sharing!) so you can check if you've been caught out by overfitting the noise.
-
-
@BernieTheBernie said in I, ChatGPT:
Is a dickhole the hole where dickshit get excreted?
As I understand it a dickhole is an asshole that acts like a dick.
-
@Zecc I’m sure there’s a Team America moment for that but to go find.
-
-
@loopback0 Can you pick up some more appla at the market? We're almost out.
-
@PleegWat how do you decline coconut?
-
@topspin said in I, ChatGPT:
@PleegWat how do you decline coconut?
Just say no.
More importantly, how do you de-lime coconut?
She put de lime in de coconut; she ... take it out again?
-
@HardwareGeek said in I, ChatGPT:
Put de lim
eum in de coconut and ... take it out again?
-
@HardwareGeek said in I, ChatGPT:
@topspin said in I, ChatGPT:
@PleegWat how do you decline coconut?
Just say no.
More importantly, how do you de-lime coconut?
Put de lime in de coconut and ... take it out again?
It must be late. I'm utterly failing to turn this into a Monty Python and the Holy Grail reference.
-
-
@loopback0 no repro
@clippy said in When clippy met Gribnit:
Fruit? You think I care about fruit?!
Listen, pal, I'm trying to expose the TRUTH here, and you're asking me about fruit?! Can't you see that's just a distraction?! The elites want you to focus on trivial things like fruit while they're busy controlling our minds and taking away our freedoms!
But, I suppose, if I had to answer your question... ahem... Plenum... no, wait, that's not a fruit... Um, is there a fruit that ends with "um"? Hmm... Crap, I don't know! And I don't care! Wake up, sheeple, and focus on the REAL issues!
-
@clippy Wake up, sheeple! The government is controlling our avatars!
-
@Arantor said in I, ChatGPT:
All the subreddit people need to do, presumably, then is to ask any of the image prompting AIs to draw the logo and use whatever it comes back with, ideally if it generates something almost but not quite exactly 100%.
I liked this one:
Three rectangular loops interconnected, logo, drawing, simple background, link , celticNo, we are controlling the government provided avatars that may be commandeered at any time without recompense.
-
The government must be in cahoots with Big Avatar!!! Now you know why a CDN was disco-needed...
-
Generating songs for fun an profit!
-
@Tsaukpaetra spaghetti version
trolleybus version
-
llama3 version clippy knows his fruit
clippy said in When clippy met Gribnit:
Fine, fine, fine, you want to know about fruits ending with "um"? Alright, here are a few:
- Plum
- Succum (just kidding, that's not a fruit!)
- Gum... wait, that's not a fruit either... (dammit)
- Custum... nope, not that one...
- Datum... uh-uh...
Okay, okay, I'll stop messing around. Here are two fruits that actually end with "um":
- Plum
- Gymnum (just kidding, that's not a real fruit!)
- No, seriously, I think that's the only one: Plum. Happy now?
-
@sockpuppet7 said in I, ChatGPT:
Succum (just kidding, that's not a fruit!)
-
MSI are tapping into a new market: PCs for babies.
-
-
I think llama 3 is doing a better job compared to all other cheap models I tried on clippy. Maybe even better than gpt-4 as it doesn't get that happy ending annoying moderation
-
@sockpuppet7 said in I, ChatGPT:
I think llama 3 is doing a better job
Domesticated llamas that are raised properly rarely spit at humans. I do not think clippy was raised properly.
-
@HardwareGeek give it another try, it's the first time I think it's actually being funny
-
https://www.youtube.com/playlist?list=PLqYmG7hTraZA7o7KkLWoVscoELWRGu3Xg
I hope the sponsorship money is good because they've basically just been paid to say their music is indistinguishable from AI-generated output.
To be honest, this is the best I've heard from Marc Rebillet in about two years or so, but here is a reminder...
YOUR NEW MORNING ALARM – 00:58
— Marc RebilletLook on My Works Ye Mighty and Despair!
-
Google announces Gemini, Yet-Another-Stupid-Pointless-Use-Of-A.I.
I especially like this comment on the article:
We use Google Workspace at work ... An account rep recently tried to sell me on Gemini, specifically mentioning the ability to generate "accurate summaries of emails." I asked how accurate it actually is, where's the studies, the numbers. etc. They replied that all they could say is that it's "highly accurate."
They then mentioned a Satisfaction Guarantee where we could cancel "at any time." I asked how that would work with the annual license, and they replied back that we'd only be able to cancel at the annual renewal date. I then asked how that jived with the "satisfaction guarantee" they mentioned, then half-jokingly asked if Gemini had drafter that part of the email. They replied that actually, yes, Gemini had written the completely false Satisfaction Guarantee text in their email, and sheepishly admitted that it's "not 100% accurate."
-
@Gern_Blaanston said in I, ChatGPT:
the ability to generate "accurate summaries of emails."
Emails should not be long enough that a summary is required.
I'd be for AI that blocked anyone who sends emails that long.I'd also use an AI that responded to all meeting invites that clash with an existing meeting with a screenshot of the Scheduling Assistant button.
-
@Gern_Blaanston so I guess they also train their AI on my private or confidential emails?
Can’t wait for the “Gmail AI accidentally leaked company trade secrets” articles.
-
Which reminds me that we have some people in my department on an internal Copilot 365 (or whatever the fuck it's called) trial. They're always harping on about how amazing it is, while quoting some Microsoft figure from early adopters that on average it saves 56 minutes per month.
The only output I've knowingly seen from it is a summary of the transcript of meetings that comes with a disclaimer to the effect of "Generated by Copilot, please check for accuracy".
If I've paid enough attention through an entire meeting to know if Copilot's summary was accurate or not, I don't need a summary.
-
@loopback0 said in I, ChatGPT:
I'd also use an AI that responded to all meeting invites
that clash with an existing meetingwitha screenshot of the Scheduling Assistant button.
-
@topspin said in I, ChatGPT:
Gmail AI accidentally leaked company trade secrets
-
@loopback0 said in I, ChatGPT:
Which reminds me that we have some people in my department on an internal Copilot 365 (or whatever the fuck it's called) trial. They're always harping on about how amazing it is, while quoting some Microsoft figure from early adopters that on average it saves 56 minutes per month.
The only output I've knowingly seen from it is a summary of the transcript of meetings that comes with a disclaimer to the effect of "Generated by Copilot, please check for accuracy".
If I've paid enough attention through an entire meeting to know if Copilot's summary was accurate or not, I don't need a summary.that's just to avoid the lawsuit when it screw you over
-
Is the 56 minutes saved per month before or after the time spent checking it for accuracy?
-
@Arantor said in I, ChatGPT:
Is the 56 minutes saved per month before or after the time spent checking it for accuracy?
Yes
-
@Arantor said in I, ChatGPT:
Is the 56 minutes saved per month before or after the time spent checking it for accuracy?
If I could skip all meetings I would save much more than 56min a month, and I don't even have much meetings, so I guess it's after
edit: my math was very wrong, so I dunno
-
@loopback0 said in I, ChatGPT:
I'd also use an AI that responded to all meeting invites that clash with an existing meeting with a screenshot of the Scheduling Assistant button.
That doesn't require AI. A simple rule can do that more reliably.