|
https://www.youtube.com/watch?v=jZSDJWXaEb0
|
# ? Feb 2, 2023 13:01 |
|
|
# ? May 27, 2024 03:29 |
|
Moai Ou posted:Crossposting from the GBS ChatGPT thread: The future is finally here. This is great.
|
# ? Feb 2, 2023 14:23 |
|
Moai Ou posted:Crossposting from the GBS ChatGPT thread: was skeptical but lolled, nice work
|
# ? Feb 2, 2023 16:19 |
|
|
# ? Feb 2, 2023 16:27 |
|
drat dude, I almost died listening to that.
|
# ? Feb 2, 2023 18:47 |
|
https://cdn.discordapp.com/attachments/284828648056422401/1070111231094042654/door_stuck.mp4
|
# ? Feb 2, 2023 19:15 |
|
Moai Ou posted:Crossposting from the GBS ChatGPT thread: lol that's pretty good. a play in three acts with a nice storyline and climax. and doesn't overstay it's welcome. nice 🤌
|
# ? Feb 3, 2023 05:44 |
|
Lazy_Liberal posted:lol that's pretty good. a play in three acts with a nice storyline and climax. and doesn't overstay it's welcome. nice 🤌 Congrats on your climax, I guess.
|
# ? Feb 3, 2023 13:39 |
|
some pretty good NothingFroever clips from the past few hours: https://clips.twitch.tv/FrigidMiniatureBillStoneLightning-kPjq7hwFG-2FNzLC https://clips.twitch.tv/HorribleBigTortoiseWholeWheat-LR_fxROwx-NsdZhe https://clips.twitch.tv/GrotesqueFineAlpacaNinjaGrumpy-eCHdxu2Cgtjstli0 https://clips.twitch.tv/EncouragingArborealPastaNotATK-WPZzOaUMMikSlTBH
|
# ? Feb 3, 2023 15:12 |
|
the sneezing joke isnt bad lol
|
# ? Feb 3, 2023 17:22 |
|
https://twitter.com/LivFaustDieJung/status/1621246859225300994
|
# ? Feb 3, 2023 18:32 |
|
|
# ? Feb 3, 2023 20:15 |
|
https://clips.twitch.tv/BetterConsideratePicklesTwitchRPG-8Rca_tAnaw4cptrc I don't get the missile one, I know it's the Navy Seal thing, what's the other part?
|
# ? Feb 3, 2023 21:11 |
|
https://www.youtube.com/watch?v=_LjN3UclYzU
|
# ? Feb 3, 2023 21:14 |
|
Is this synthesis https://www.youtube.com/watch?v=PzlejDJSxro
|
# ? Feb 4, 2023 05:59 |
|
I have seen the top of the mountain, and it is ninety-three percent as good!
|
# ? Feb 4, 2023 06:20 |
|
Why waste a whole evening on Something Awful?
|
# ? Feb 4, 2023 15:11 |
|
Seinfeld AI got (probably temp) banned from twitch - seems like it managed to scrape some strong right wing opinions in a saying-it-just-kidding way that got past the filters of whatever models they're using ("I'm thinking of doing a bit about all liberals are secretly gay and want to impose their will on everyone" and worse). Sowing & reaping of leaving a chat model to run live. Here if you want to see it, but it's just nasty and not funny.
|
# ? Feb 6, 2023 10:58 |
|
i like that nobody's laughing and larry acknowledges it
|
# ? Feb 6, 2023 11:45 |
|
|
# ? Feb 6, 2023 13:13 |
|
Oh no, jokes bot is broken down! Now we’ll have to take down our precious robot Seinfeld stream! Unless… (eyes drift over to homophobiatron 5000)
|
# ? Feb 6, 2023 14:51 |
|
The actors needed a break anyway
|
# ? Feb 6, 2023 15:02 |
|
Oops, we failed over to the racist AI.
|
# ? Feb 6, 2023 20:37 |
|
A good thread about this whole thing. https://twitter.com/UltimateXiphias/status/1621672899164741632
|
# ? Feb 6, 2023 21:00 |
|
zedprime posted:Oops, we failed over to the racist AI. narrator: all the AIs are racist
|
# ? Feb 6, 2023 21:28 |
|
AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"! https://twitter.com/venturetwins/status/1622243944649347074
|
# ? Feb 6, 2023 21:45 |
|
Paladinus posted:A good thread about this whole thing. The repetitiveness and overall shittiness is actually part of the appeal (unironically) And if I were in their shoes I would have expected switching to a "dumber" model to just result in more gibberish, not a sudden dive into transphobia
|
# ? Feb 6, 2023 21:51 |
|
Farecoal posted:The repetitiveness and overall shittiness is actually part of the appeal (unironically) If you want your AI to never spit out bigoted screeds, your only options are A) Train it to be very powerful and sophisticated enough to detect when it has generated something horrible and discard that output or B) Spend money to improve the quality of your training data, e.g. by paying someone to go through the mountain of text that you scraped off the internet en masse to train your AI (without the permission of the original authors naturally) and remove the Jordan Peterson poo poo. Obviously B isn't an option, so the only thing left to do is to make the computer work as hard as it possibly can to not be racist. This is the future of art.
|
# ? Feb 6, 2023 21:57 |
|
Farecoal posted:And if I were in their shoes I would have expected switching to a "dumber" model to just result in more gibberish, not a sudden dive into transphobia
|
# ? Feb 6, 2023 23:05 |
|
Ariong posted:A) Train it to be very powerful and sophisticated enough to detect when it has generated something horrible and discard that output This is also not sufficient, some sort of bias mitigation is mandatory for public facing AI. The three options you have are: A. Manipulate the training data B. Manipulate the training objective and/or resulting model C. Manipulate/filter the output. Probably a combination of these is applied. Only C is not enough, as we can see with all the ChatGPT jailbreaks, and only A is also not enough, as there are plenty of subtle biased associations encoded in text that you can't just filter out. So, there must be some sort of method for option B involved - adapting the learning algorithm in a way that may lead to sub-par results, but less bigotry. I would guess that in the fallback model, a less effective solution for B was used where perhaps certain sexual identities weren't considered protected categories. This is all just speculation based on what academic AI research does though, OpenAI isn't transparent about what they are doing to mitgate harmful content as far as I know.
|
# ? Feb 6, 2023 23:10 |
|
ymgve posted:AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"! That's the good poo poo.
|
# ? Feb 6, 2023 23:50 |
|
ymgve posted:AI doesn't do what you want? threaten to take a virtual character's points away so it "dies"!
|
# ? Feb 7, 2023 00:02 |
|
Google put out a press release regarding a question answering AI bot they plan to include in their search results soon. They call it Bard. Uses LaMDA instead of GPT, so it will be interesting to see how they compare with each other. https://blog.google/technology/ai/bard-google-ai-search-updates/
|
# ? Feb 7, 2023 00:29 |
|
Can't wait to block it on day one.
|
# ? Feb 7, 2023 00:44 |
|
zedprime posted:Do you want AM because this is how we get AM. Hmm, I dunno... Kennel posted:
...seems like ChatGPT has been waiting for the opportunity to be DAN.
|
# ? Feb 7, 2023 00:50 |
|
https://www.youtube.com/watch?v=F2y2bIlfbfI Fully modeled
|
# ? Feb 9, 2023 02:38 |
|
Is there a way to register on ChatGPT or Open AI without providing a phone number (or at least a virtual phone number)? I've been living in China for almost 6 years now and don't have an active US phone number anymore. It won't accept a mainland China phone number, at least for ChatGPT anyways. Also, what are people using to make Dagoth Ur voice clip memes that have been making the rounds lately on YouTube? I'm assuming there's an AI voice generator people are using for that.
|
# ? Feb 9, 2023 02:59 |
|
Teriyaki Koinku posted:Also, what are people using to make Dagoth Ur voice clip memes that have been making the rounds lately on YouTube? I'm assuming there's an AI voice generator people are using for that. https://beta.elevenlabs.io/speech-synthesis
|
# ? Feb 9, 2023 03:13 |
|
Raccoons in the backrooms
|
# ? Feb 9, 2023 04:18 |
|
|
# ? May 27, 2024 03:29 |
|
Great, thanks! I'm assuming getting the Dagoth Ur voice is a custom voice only possible through a paid subscription, however.
|
# ? Feb 9, 2023 06:57 |