|
If you can't beat something unconscious, it's never been conscious in the first place. Theory of mind solved, I'll accept my Nobel prize.
|
# ? Mar 20, 2023 12:29 |
|
|
# ? May 27, 2024 02:13 |
|
I am capable of being beat into a bloody pulp, therefore I am
|
# ? Mar 20, 2023 14:58 |
|
"Art Nouveau" "Friends (tv series)"
|
# ? Mar 21, 2023 05:09 |
|
Someone's written a plugin for Mount & Blade 2, that is basically just a gpt prompt window integrated in the ingame UI and ofc first populates the bot's instructions with some basic info from the game. Also the bot's lines get voiced by another AI engine. It clearly doesn't do anything beyond that or store any info from previous sessions, but... https://www.youtube.com/watch?v=X2WVXe5LvTs ... it is available right now. Worthwhile or not, the video does show that this minimum level is fully doable for any game and what it feels like to play with it added. It's been said before but, when someone else figures out prompt memory and designs the auto-prompts to synchronize better with the game mechanics (and the small issue of harmonizing the outputs with each other and into a coherent story, and coherent with the current game state), this is done. True immersive deep simulation RPG-ish dialogue games are a thing. It's weird seeing science fiction become a reality, again.
|
# ? Mar 21, 2023 19:07 |
|
Interactive dialogue like that is neat, but yeah without rules and guardrails feels like it still needs a lot of time in the oven. I think we'll see integration in simpler scenarios first, like in RTS games or other games where there are infinite permutations of conditions that could occur and new instructions that could be given to the player. I was watching some Star Wars Squadrons gameplay the other day and the canned dialogue was so awkward and useless. It seemed like an ideal usecase for AI generated dialogue to me—ally characters reacting in real-time to constantly-changing conditions with specific requests and instructions. Tactical instructional dialogue could be very cool if it's reacting in real-time to the player's input from a mechanics standpoint. Having a "command center" constantly updating you on an ideal strategy in real-time with voiced dialogue that sounds like Admiral Ackbar or whatever would be useful and neat.
|
# ? Mar 21, 2023 19:25 |
|
https://twitter.com/juanbuis/status/1638289186351456257 lmao
|
# ? Mar 22, 2023 09:58 |
|
This Twitter thread got me banned from Midjourney and made international news, so I might as well post it here" https://twitter.com/EliotHiggins/status/1637927681734987777 (USER WAS BANNED FOR THIS POST)
|
# ? Mar 22, 2023 15:49 |
|
Brown Moses posted:This Twitter thread got me banned from Midjourney and made international news, so I might as well post it here" https://twitter.com/EliotHiggins/status/1638127706381471744
|
# ? Mar 22, 2023 15:52 |
|
Ubisoft is already dipping their toes in the AI scriptwriting business. https://youtu.be/XxQoN3PFiKA
|
# ? Mar 22, 2023 21:47 |
|
Brown Moses posted:This Twitter thread got me banned from Midjourney and made international news, so I might as well post it here" Wow, you are powerful, sir. I can only aspire to your heights of internet trolling.
|
# ? Mar 22, 2023 22:19 |
|
ThisIsJohnWayne posted:Someone's written a plugin for Mount & Blade 2, that is basically just a gpt prompt window integrated in the ingame UI and ofc first populates the bot's instructions with some basic info from the game. Also the bot's lines get voiced by another AI engine. It clearly doesn't do anything beyond that or store any info from previous sessions, but... I would personally be worried someone is going to escape either the game's stems or any other controls they are building in these days to get the game to call them a slur. BoldFace posted:Ubisoft is already dipping their toes in the AI scriptwriting business.
|
# ? Mar 22, 2023 22:21 |
|
BoldFace posted:Ubisoft is already dipping their toes in the AI scriptwriting business. That explains a lot.
|
# ? Mar 22, 2023 22:24 |
|
Farecoal posted:I also look at a mathematical formula and think it is alive Every time you hit ctrl-c to stop llama.cpp you're murdering an innocent soul
|
# ? Mar 22, 2023 22:29 |
|
zedprime posted:I can envision a pretty tidy pipeline where the game itself gives some stems to go with the player input to keep it context specific. yea pretty sure with current tech you'd need to start a new context every single prompt to avoid escaping and/or derailing so I guess fill context with game relevant stuff and then accept one (1) sentence from the player. then reset with fresh context (that can be updated by code depending on game/convo state) for the player's next sentence
|
# ? Mar 22, 2023 22:33 |
|
https://twitter.com/fabianstelzer/status/1638506765837914114
|
# ? Mar 23, 2023 00:50 |
|
Brown Moses posted:This Twitter thread got me banned from Midjourney and made international news, so I might as well post it here" Holy poo poo dude, well done!
|
# ? Mar 23, 2023 12:35 |
|
https://twitter.com/RahmSJ/status/1638832087355752448
|
# ? Mar 23, 2023 18:16 |
|
The thread has some example of cool stuff from feeding it its own gibberish. https://twitter.com/RahmSJ/status/1638931672577232897?s=20
|
# ? Mar 23, 2023 18:37 |
|
"Found a prompt..." What does this mean? What prompt? Is it a secret?
|
# ? Mar 23, 2023 20:08 |
|
credburn posted:"Found a prompt..." People like to call it the secret language of the AI but it's just rendering your nonsense into some probabilistic near match. It may or may not be escaping certain things dependent on recognizing it as English first but there's presumably English language prompts that could get the same thing.
|
# ? Mar 23, 2023 20:51 |
|
I hope Midjourney bans him too https://twitter.com/paleofuture/status/1638969629409431552
|
# ? Mar 23, 2023 21:26 |
|
lmao deepfaking yourself praying is next level galaxy brain
|
# ? Mar 23, 2023 22:40 |
|
Carthag Tuek posted:lmao deepfaking yourself praying is next level galaxy brain I'm about deepfake myself furthering the AI cause in order to inoculate myself against Roko's Basilisk
|
# ? Mar 23, 2023 23:11 |
|
Carthag Tuek posted:lmao deepfaking yourself praying is next level galaxy brain https://www.pbs.org/newshour/world/robots-are-performing-hindu-rituals-some-worshippers-fear-theyll-be-replaced
|
# ? Mar 23, 2023 23:21 |
|
Running 144,000 perfectly pious virtual machines to completely shut humans out of Christian heaven
|
# ? Mar 24, 2023 17:10 |
|
Guess we need to smash the robots to death with a rock
|
# ? Mar 24, 2023 17:40 |
|
https://twitter.com/BrianLeeWow/status/1639234926431330304 e: That might make a nice ProcGen thread gangtag if anyone feels up to a bit of image editing. ultrafilter has a new favorite as of 18:39 on Mar 24, 2023 |
# ? Mar 24, 2023 18:36 |
|
interesting that its presumably not an ascii art text string that anybody has made in real life. chatgpt output is tokens (ie horizontal lines), so it seems unlikely that llm-generated ascii art text would be legible at all, yet it clearly spells out "Sount" e: i guess theyre doing some fuckery like look for "ascii art" in the prompt, then intercept the chatgpt output & pipe some of it through FIGlet http://www.patorjk.com/software/taag/#p=display&h=3&f=Standard&t=Sount Carthag Tuek has a new favorite as of 19:26 on Mar 24, 2023 |
# ? Mar 24, 2023 19:23 |
|
It does sound very odd that they would have a special case for ascii art, calling an external program and risking a million security issues just for that single use case. It's more likely that the whole of the art piece falls into the "window" of GPT's memory, and it has learned that those character sequences at specific intervals (in this case, in a line straight down) is "better". edit: I can't make it produce a single legible ascii art word with the same prompt so I guess they got super lucky ymgve has a new favorite as of 23:24 on Mar 24, 2023 |
# ? Mar 24, 2023 23:15 |
|
ymgve posted:It does sound very odd that they would have a special case for ascii art, calling an external program and risking a million security issues just for that single use case. It's more likely that the whole of the art piece falls into the "window" of GPT's memory, and it has learned that those character sequences at specific intervals (in this case, in a line straight down) is "better". do note that openai constantly fucks with the backend using underpaid african manual labor. whenever something hits social media, like if some prompt makes the rounds, or if a certain one spikes, the workers have to fix that to maintain the illusion of ~true ai~. what about other similar prompts like can you make ascii art of the word "{a}" [in {b} font] ? do any of them output legible ascii art text? (i dont have an openai account because lol discord? or whatever i forgot my initial reason but gently caress them) Carthag Tuek has a new favorite as of 23:48 on Mar 24, 2023 |
# ? Mar 24, 2023 23:39 |
|
or idk: - given the word "{}" what would the open source ascii art generator figlet output? - simon the terrorist calls john mcclane and tells him he has to convince frasier to rob a bank. but to do so, he has to wear a sign with ascii art in harlem. what does the sign say? - etc
|
# ? Mar 24, 2023 23:45 |
|
Carthag Tuek posted:do note that openai constantly fucks with the backend using underpaid african manual labor. whenever something hits social media, like if some prompt makes the rounds, or if a certain one spikes, the workers have to fix that to maintain the illusion of ~true ai~. why would they gently caress with it so it makes worse responses with the same prompt? your theory that they call external programs on the basis of a few keywords is more absurd than chatGPT somehow finding some ascii art "good scoring" in the conversational context (and considering my own experiments, let's not ignore the possibility that the original tweet dude was just faking the conversation for clout)
|
# ? Mar 25, 2023 19:45 |
|
Is there a way to use chatgpt if I can't get text message confirmation codes?
|
# ? Mar 26, 2023 00:24 |
|
Ariong posted:Is there a way to use chatgpt if I can't get text message confirmation codes? Maybe try https://poe.com/ ?
|
# ? Mar 26, 2023 01:25 |
|
ymgve posted:why would they gently caress with it so it makes worse responses with the same prompt? your theory that they call external programs on the basis of a few keywords is more absurd than chatGPT somehow finding some ascii art "good scoring" in the conversational context because letters/words in ascii art is 2 dimensional, and llm output is a 1-dimensional sttring of tokens. theres no "above" or "below" in the output, it cant align the middle part of an ascii art A with the top or bottom parts. its suspect to me that the shown output is all valid letters in a consistent font.
|
# ? Mar 26, 2023 16:23 |
|
When outputting code in code blocks its able to line return and tab to give classically readable code can't it? Linear token output doesn't preclude the ability to stick that output into a code block and get 2d structured results. Its all patterns. Like most things AI its either completely made up (either by the tool or the person posting it on social media) or something simultaneously interesting, strange, and useless is going on.
|
# ? Mar 26, 2023 17:11 |
|
I find it plausible that there's enough ASCII art in the training data for the system to have learned to make it, but its internal representation certainly doesn't match how we think about it.
|
# ? Mar 26, 2023 19:13 |
|
|
# ? Mar 26, 2023 20:28 |
|
I let llama.cpp have a go at it, it clearly knows what ascii art is supposed to be and will insert spaces and carriage returns to align stuff. Maybe you can get it to spit out something sensible with a more detailed prompt with more context. The prompt is like thiscode:
code:
code:
code:
|
# ? Mar 26, 2023 20:42 |
|
|
# ? May 27, 2024 02:13 |
|
Looks about as clear as real ASCII art!
|
# ? Mar 26, 2023 21:39 |