Plus the fact that Duncan was the traitor who swapped with Kirsty means he should have known she was bad, idiots all of them (jk it's probably much harder to keep track of things when you're actually playing)
@pavonian7531 In actual TTT the issue isn't that finding the T is hard. People usually figure it out quickly, but unfortunately for them it often winds up being a bunny hopping maniac on tapping you with a deagle.
You're the last Innocent alive. The traitors have just shot the Detective, nearly finishing their ruthless slaughter of the Innocent team. You're almost out of ammo, and you can hear footsteps approaching from down the hall. Are you about to be the final victim of this massacre in Terrorist Town, or are you on a... Mystery Quest!
I seem to recall that Rythian's strategy is half the reason the Hypnotist exists, so that you can't automatically trust the "I'll rez them after you kill to check!"
Something about this map is stirring a long-forgotten memory. In the deepest recesses of my mind, a sassy voice that sounds like Lewis speaks. I think I can hear what it says: *_zOeY_* *_iS_* *_a_* *_sH!t_* *_cAt_*
@@SyntheticSpy This made me laugh, and then realize that yeah. either him, or Athena. Maybe Apollo or Artimis though, considering their use in hunting. Hestia too, when used for home defense. Ok a lot of gods would have guns in their jurisdiction, depending on what they are used for. Maybe that's why I can't think of a single god who has sole Jurisdiction over a weapon type.
I can’t believe Rythian was genuinely trying the ol’ “kill and res” strat *in a role pack with a hypnotist*. I was 100% certain he was bad and joking around when he first tried that lmao
The gonk droid was made out of a silver reprocessing unit. Its a machine that takes the silver out of exhausted photo fixer. I'm certain a lot of props from star wars are just pieces from a darkroom George Lucas had access to.
@@CaseChronicles69 Hahaha. Well, then allow me to provide some things to improve your skills of deduction on that front then - since you'd *never* expect an LLM-based AI to make a comment like that OR sound that way when doing so (and it's a very good skill to actually be able to tell the difference between human & AI posts - likely even moreso than early Internet fact checking was). First, my comment is referencing an obscure moment from an old Yogs TTT video played on this same map where Osie's sister Bouphe was acting like her in this video by obsessing over a Star Wars "droid" which is something an AI wouldn't pick up on. Second, making use of _italic_ markup to convey tonal sarcasm rather than reinforcing the contextual emphasis of _“definitely complete”_ is a nod to a very specific detail from the old video that isn't obvious in plain text (which is that Bouphe's "droid" was a just disconnected foot from an R2 unit) which is also not something you'd expect from an LLM AI response. Third, even if an LLM AI scrapes the auto captions of all the Yogs videos, you'd probably get details like the word "Gonk Droid" in plain text rather than using the Star Wars designation GNK, again because that's a detail that you can only accurately alter when you have the appropriate underlying context of knowing exactly what and how that word is used. Fourth, UA-cam bots stopped using timestamps when those comments didn't just boost the likelihood of user engagement & exposure to get people to click their booty profile links… and even when they still did they never had comments that were contextually relevant to the timestamp provided, because they're made with no understanding of the specific video content to be able to do that. LLMs are just applied statistics, they're not really an artificial intelligence of any kind, so it's always best to think of them that way. This is why they're incapable of assembling comments like my OP - because they have a complete abstraction to the underlying context of the words they're just probabilistically stringing together to form a likely valid/correct reply. This is why AI and other UA-cam bots still only make _extremely generic_ comments on videos. It's less about how verbose or what manner of speaking someone used in a post online that will tip you off one way or another (as that can vary heavily with literacy or whether or not you're using your primary language). It's more about how much you can discern the presence of small, but simple details that LLM AI can't discern - but which humans will _almost never_ misunderstand. An easy example to remember is that you could tell a small child, _“The trophy wouldn't fit into the suitcase because _*_it_*_ was too big.”_ and the child knows that the "it" that was too big is the trophy. An LLM AI has no way to determine if the "it" is the trophy or the suitcase that's too big, because it doesn't know what any of those things are. This is the same reason why a child can tell when a human hand looks wrong but LLM prompt generated AI Art constantly struggles with correctly representing those types of details even in still images, let alone in video. It's also why AI Art and AI text in articles spreading online so rapidly is also creating an "average" that's likely also polluting future LLM AI by essentially recursively feeding them their own output as equivalent to human-generated content. LLMs can be very specific contextually… but *only* if & when they have a small or VERY consistent dataset of how humans respond to a given prompt (and even then they mix & match to make an original answer that loses context and easily creates factually inaccurate comments). That's why video comments made with context that requires visual, pre-existing relationship, and historical references to understand in order to make them are VERY unlikely to ever be AI generated. On webforums, early spambots started making patchwork posts assembled from other peoples' replies to make it harder to identify, but they still got all that context from the structural layout of how the subforums & threads are configured to predict with higher probability what a post that was likely to be considered "human" is and not caught by auto-moderation tools. That's a contributing factor to why I have a very… distinctly tl;dr posting style - because it's very obvious when any section of what I say is being copied & reposted elsewhere. I've been modding forums for the last 20 years and thus seeing how spambots "fake" human-ness has gradually improved over time up to the big LLMs that are everywhere now, but it is VERY EASY to spot an AI if you know what to look for in what they're unlikely to be able to do well vs. what humans do all the time usually without even thinking about it. It's also why I find it helpful to just drop information most people are unlikely to have been exposed to when I see simple mistakes like that. Hopefully now you'll be a _little_ better equipped to know what is or isn't an AI post the next time you're browsing UA-cam comments. (Since, despite your very auto-generated, burner account-looking username, your comment on this video about Tom's "return" and reaction to Angor-bot 5000 still rings quite human and provides context on why _you're_ likely to have responded with an implied familiarity with _my_ commenting style in particular). Cheers!!
@@Cyrathil Increasing the size of the data set that an LLM uses doesn't change _anything_ I mentioned. They're still just applied statistics and not actual artificial intelligence. A bigger data set doesn't provide it any ability to have real analytical context, and that nuance becomes the critical breakpoint as more very VERY small contextual nuances will continue to radically alter if and how an answer is valid or not - which is why poking at those weaknesses makes it easy to see when something is an AI and when it isn't. When you're looking at a system that is vast & complex AND that is extremely sensitive to initial conditions, this is also known as being deterministically chaotic. Predictive models are GREAT for those which is why LLMs are genuinely exciting and interesting… but they also exist with some _very_ hard limitations on how specifically accurate they can ever be, _LITERALLY no matter how much data you add to them._ This is the same reason that weather (the classic deterministically chaotic system along with the three-body-problem) doesn't have a predictive system that remains valid past roughly a 10-day forecast no matter how enormous the quantity of data that you feed into it is. Outside of that window the most accurate predictive models just become the statistical average of the weather on that day over the full historical record… but that's still not the truth of what the weather actually is when it happens, and why am LLM attempting to provide small nuanced answers has core weaknesses it can't escape. The difference here is that these bits of small sensitive differentiation in AI chat bots are from human context… _which LLMs don't and can't have, but which humans ALWAYS have because that's how our brains work._ Our brains do the same sort of predictive things LLMs do when we copy something we've seen in order to test a pattern to try and understand something we don't know and then learn the context about that thing… Without that second part, it can't get any better past a certain point. That's why LLMs are not ACTUALLY an AI and why I describe them as applied statistics to help highlight that. As an easy real world human parallel, if your 4-year-old sibling suddenly swears at the dinner table by just repeating a word they heard - they don't know what it is, they don't know what it means, but they do understand a likely pattern it's been used in and can mimic that to seem like they do. When I went to Germany when I was about 5 or 6, after a couple days I easily responded to someone saying, _“Wie geht's?”_ to me with, _“Mir geht's gut, un dir?”_ but I didn't know what ANY of EITHER sentence meant because I didn't speak German. I just was around it often enough that I knew that's how conversation flowed, but those words were empty of ANY context. That's ALL an LLM ever does, but with the entirety of its data set being exponentially larger. Nothing they assemble has any context for what those words are or what about them specifically changes how they're used. It's very easy for people to assume it can because that's what we expect that conversation to have behind it when it seems correct enough… but LLMs don't have that at all. They're just outputting whether or not it's statistically likely to work and they can't do more than that. The difference is that the 4-year-old does that to learn what that swear word is and when to use it and when not to based upon human social context in which they as a single individual exist. The mirror neurons that let you copy behaviour like that are just one VERY SMALL region of the front of the prefrontal cortex. Context for each individual also changes over time, in which group you're in (friends vs. family vs. school vs. church vs. online community vs. cultural background etc.) which alters how or when that can be used specific to tiny details that AI have no context to . Those are all human contexts we can't exist without because of how our brains work and are designed for socialization as a facet of survival… and that LLMs can't have because of what they are and all the things they don't have beyond just a massive dataset. LLMs don't suddenly get better at doing ANY of the things I mentioned in my initial reply even as they do get statistically more probable at generating something that passes for a valid answer more often - like Chat GPT 3 to 4 apparently did, just like we got slightly better weather forecasting. LLMs still don't have any analytical context, and will still have those same weaknesses when you know what to look for that makes them stand out.
@@PierceArner The size of a LLM isn't just the size of its input. It's the size of the model internally, the number of nodes making connections. That's implicitly tied to the depth of connections it can make and from 3 to 4 it absolutely has started making analytical connections. A recent example I saw was physical understanding. If you ask 3 to stack items it'll jumble them together because that's not a common task in its corpus and it's not got an awareness of physics to give an at all decent shot. 4, while not perfect, is actually able to give a passing shot. There's no guarantee that the trend will continue, but I think a lot of people are discounting emergence as a principle despite pretty much all of our natural features arising out of emergence.
A note about Gonk Droids: they have always existed and had a purpose, they weren't a mistake or unknown addition. They were originally labelled as "Power Droids" (via an official collectible toy) and were designed to be walking batteries - extremely simple as far as AI functionality goes, just having enough to know how to navigate and go where they're needed. However soon after they were revealed in the movies news sites and fans latched onto the noise they made and called them "Gonk" Droids. This ended up becoming official, and they got retconned into "GNK" series power droids, commonly referred to as Gonk Droids by the wider galaxy.
The reason there is a molten pit of metal on the sandcrawler is because they started out as a mobile refinery. They subsequenctly got abandoned when the ore ran out (or there wasn't any ore to find, I dunno about the exact reason) and the Jawa took over and started selling droids out of them. And melting down crashed spaceships they happen to find in the desert. Citation needed.
10:00 That would be have been so funny if that worked! LOL 13:45 Gosh the swapper adds so much chaos and I love it. LMAO!! At the end there, Nilsey and Kirsty CLEANED THE FLOOR with everyone until Duncan came in. Ha!
I have an idea for a role. It's called Synth or Copy or something like that. What it does is you scan a target and then can create really shitty bot clones of that person that attack anyone not on the same team as the creator. Like a summoner but one that sows discord because you make say a fake Kirsty and it starts running around making people think Kirsty is bad. They'd be super bad at aiming (so for most yogs that would blend in better lolol) so they're not great, but they'd be able to make maybe one of each player and it could get real confusing and messy. Idk the mechanics so if creating bots can't work easily and in that many numbers, maybe it's crappy combine soldiers retextured as the player whose dna you stole? Something like that.
One of the things I love about the original Star Wars was how cobbled together the production was. It was part of it's charm. Two of the lightsabers used in the first movie were used camera flash handles that the propmaster found at a used electronics place. Then he just added some extra parts on it, including fins from the windscreen wipers on his truck. Obi Wan's lightsaber was partly the same thing with a WW1 rifle grenade added.
Got a fun idea for how the Assassin could work: the ability to toggle on or off their damage. By default they’re a regular traitor, however a unique item would let them deal no damage until toggled back on. Then they can “out themselves as a jester”, make people let their guard down, then harpoon em in the back. This would also buff jesters by association, seeing how every jester could be an assassin.
I've been playing with the map editor and have considered making a map in my spare time for fun, it's a really cool program. Maybe if I'm lucky they'll try it out I know they're always looking for more. If anyone has any ideas for stuff to add in lmk. It likely won't be finished any time soon and it probably won't be amazing but it seems like a fun little side thing to try out. I'm better with functional stuff not so much visual design, but I'm not amazing with either so keep that in mind if you do have any suggestions lmao
Does anybody else have a problem with how often the Death Note gets used in 1v1s? It's a lose button for the traitors like 90% of the time, probably statistically actually higher.
I wish they would update the Tom-bot. Remember when he killed everyone on the map with a free kill gun that one time? Code him up to be a moving sniper with a free kill gun that shoots any role that isn't the one that spawned him... and give him all of Toms creepiest lines from every recording he's been in.
Didn't Rythian killing Nilsey actually prove he was innocent? If he was a traitor he would've known that Nilsey was some sort of jester role and wouldn't have taken the chance killing him like that
Rythian’s strat is shit as a paramedic but amazing if you’re a hypnotist since you can kill an innocent, get the detective to confirm them, then res them as a traitor
Rhythian's "Everyone should want to be seen, unless you're a traitor. because its good to be tracked." is legitimately scary sentiment: nothing to hide, nothing to fear.
Great video like always! If I could suggest a game you guys should check out colony survival it’s added lots of new stuff since last time you guys played it! :)
@@Home-u6g everysingle one? you mean this single comment? LMFAOOOOOOO also how is saing "a gonk droid is not a walking trashcan" talking shit? LMFAOOOOOO you must be prosocial af. you see someone talking about a droid form star wars and you think that person is talking shit about your favorait UA-camr LMFAOOOOOOOOOOOOOOOO stay triggered little snowflake
What would happen if someone used the deathnote then killed the swapper? would they die anyway because it's a forced death or would they live because killing the swapper (atleast used to) counts as dying then coming back.
every time the last traitor forgets theres a glitch, "wait youre not bad?" you love to see it
Plus the fact that Duncan was the traitor who swapped with Kirsty means he should have known she was bad, idiots all of them (jk it's probably much harder to keep track of things when you're actually playing)
@pavonian7531 In actual TTT the issue isn't that finding the T is hard. People usually figure it out quickly, but unfortunately for them it often winds up being a bunny hopping maniac on tapping you with a deagle.
You're the last Innocent alive. The traitors have just shot the Detective, nearly finishing their ruthless slaughter of the Innocent team. You're almost out of ammo, and you can hear footsteps approaching from down the hall. Are you about to be the final victim of this massacre in Terrorist Town, or are you on a...
Mystery Quest!
I seem to recall that Rythian's strategy is half the reason the Hypnotist exists, so that you can't automatically trust the "I'll rez them after you kill to check!"
Nilesy: "Kirsty!? As I live and breathe ... Kirsty"
Kirsty: "Not for long"
lmao
Something about this map is stirring a long-forgotten memory. In the deepest recesses of my mind, a sassy voice that sounds like Lewis speaks. I think I can hear what it says:
*_zOeY_* *_iS_* *_a_* *_sH!t_* *_cAt_*
Man, now I wanna watch the episode again, do you remember the title ?
Bouphe calling an Astrometric droid's foot a puppy comes to mind.
I must know as well.
For me its achievement hunter and now im sad
Bame Blen
"whats ares the god of again?"
"shit guns"
this made me laugh way too hard
I feel like that would actually fall under his jurisdiction
@@SyntheticSpy This made me laugh, and then realize that yeah. either him, or Athena.
Maybe Apollo or Artimis though, considering their use in hunting.
Hestia too, when used for home defense.
Ok a lot of gods would have guns in their jurisdiction, depending on what they are used for. Maybe that's why I can't think of a single god who has sole Jurisdiction over a weapon type.
I can’t believe Rythian was genuinely trying the ol’ “kill and res” strat *in a role pack with a hypnotist*. I was 100% certain he was bad and joking around when he first tried that lmao
he can't stop meta-gaming..
The episode in which people find it impossible to actually say the vital words "I am in a barnacle" when they are, in fact, in a barnacle.
"Are you not bad" him calling out ose having a death note should've tipped her off but the moment when they both realise was quite hilarious
There's just something so comedic about how quick Osie was to open fire on Duncan
10:00
I mean most people would if they were being pushed into a smelting pit.
Wish ben and zoey were here to teach us new star wars lore facts like lum guzzling.
+
But you know they would get distracted hosting another episode of the “Charlie and the Chocolate Factory” Podcast
@@Muongoing.97c so true. 🤣
Listening to some Jizz music with Max Rebo and Sly Snoodles
The physician's role sounds like an ability the medic should have rather than a role tbh
Both roles are individually powerful.
In most TTT groups physician would be decent but Yogs are usually too incompetent to pull of sneaky plays anyways so it's just kinda pointless.
The issue with new "health-based" roles is all the synonyms are taken.
@@LoganChristiansonI'm waiting for the dietician role to get released
@@iamnotjcook Don't tempt me. I'll freaking do it.
somehow... tom has returned.
And he brought his achievement with him.
I love Rythian bringing back the classic move of killing someone and then resurrecting them if they're good
aka rdm
@@TomBombadil851 Rythian Doin' Murders! A classic.
It doesn't actually work though since the hypnotist is in the pack.
@@MarkoPolo1223 It doesn't work for the rest of the innocents maybe but it does work for the paramedic
@@Envyre_ the jester/swapper AKA the anti-RDM mechanic exists. They basically say F that strat. I was waiting for Rhythian to try to kill one
The gonk droid was made out of a silver reprocessing unit. Its a machine that takes the silver out of exhausted photo fixer. I'm certain a lot of props from star wars are just pieces from a darkroom George Lucas had access to.
Obi Wan’s lightsabre was made of a flash module from a camera. Your theory is checking out so far.
@@jb-br8bf Ironically the flash gun is more valuable then graflex cameras because of this.
@@CammanderDart Makes sense. Nowadays Acer KB101 keyboards are worth more than the computers they were attached to for a similar reason.
Qui-Gon's communicator in Phantom Menace was based on Gillette razor for women.
The nearly lethargic "wait I'm Nilesy" really got me good, that was very funny.
"What's Ares the god of again?" "Shit guns."
ol needledick makes me miss tom so much 🥲
Rythain's strategy is so far a 2 for 2. First he picks the clown and then a traitor. He might be onto something
Wow Duncan’s legendary comeback happened
Duncan’s Back!
Aw, fond memories of Bouphe discovering her adorable _definitely complete_ little droid friend on this map when Osie finds the GNK at 2:16
@@CaseChronicles69 Hahaha. Well, then allow me to provide some things to improve your skills of deduction on that front then - since you'd *never* expect an LLM-based AI to make a comment like that OR sound that way when doing so (and it's a very good skill to actually be able to tell the difference between human & AI posts - likely even moreso than early Internet fact checking was).
First, my comment is referencing an obscure moment from an old Yogs TTT video played on this same map where Osie's sister Bouphe was acting like her in this video by obsessing over a Star Wars "droid" which is something an AI wouldn't pick up on.
Second, making use of _italic_ markup to convey tonal sarcasm rather than reinforcing the contextual emphasis of _“definitely complete”_ is a nod to a very specific detail from the old video that isn't obvious in plain text (which is that Bouphe's "droid" was a just disconnected foot from an R2 unit) which is also not something you'd expect from an LLM AI response.
Third, even if an LLM AI scrapes the auto captions of all the Yogs videos, you'd probably get details like the word "Gonk Droid" in plain text rather than using the Star Wars designation GNK, again because that's a detail that you can only accurately alter when you have the appropriate underlying context of knowing exactly what and how that word is used.
Fourth, UA-cam bots stopped using timestamps when those comments didn't just boost the likelihood of user engagement & exposure to get people to click their booty profile links… and even when they still did they never had comments that were contextually relevant to the timestamp provided, because they're made with no understanding of the specific video content to be able to do that.
LLMs are just applied statistics, they're not really an artificial intelligence of any kind, so it's always best to think of them that way. This is why they're incapable of assembling comments like my OP - because they have a complete abstraction to the underlying context of the words they're just probabilistically stringing together to form a likely valid/correct reply.
This is why AI and other UA-cam bots still only make _extremely generic_ comments on videos. It's less about how verbose or what manner of speaking someone used in a post online that will tip you off one way or another (as that can vary heavily with literacy or whether or not you're using your primary language). It's more about how much you can discern the presence of small, but simple details that LLM AI can't discern - but which humans will _almost never_ misunderstand.
An easy example to remember is that you could tell a small child, _“The trophy wouldn't fit into the suitcase because _*_it_*_ was too big.”_ and the child knows that the "it" that was too big is the trophy. An LLM AI has no way to determine if the "it" is the trophy or the suitcase that's too big, because it doesn't know what any of those things are.
This is the same reason why a child can tell when a human hand looks wrong but LLM prompt generated AI Art constantly struggles with correctly representing those types of details even in still images, let alone in video. It's also why AI Art and AI text in articles spreading online so rapidly is also creating an "average" that's likely also polluting future LLM AI by essentially recursively feeding them their own output as equivalent to human-generated content.
LLMs can be very specific contextually… but *only* if & when they have a small or VERY consistent dataset of how humans respond to a given prompt (and even then they mix & match to make an original answer that loses context and easily creates factually inaccurate comments). That's why video comments made with context that requires visual, pre-existing relationship, and historical references to understand in order to make them are VERY unlikely to ever be AI generated.
On webforums, early spambots started making patchwork posts assembled from other peoples' replies to make it harder to identify, but they still got all that context from the structural layout of how the subforums & threads are configured to predict with higher probability what a post that was likely to be considered "human" is and not caught by auto-moderation tools. That's a contributing factor to why I have a very… distinctly tl;dr posting style - because it's very obvious when any section of what I say is being copied & reposted elsewhere.
I've been modding forums for the last 20 years and thus seeing how spambots "fake" human-ness has gradually improved over time up to the big LLMs that are everywhere now, but it is VERY EASY to spot an AI if you know what to look for in what they're unlikely to be able to do well vs. what humans do all the time usually without even thinking about it. It's also why I find it helpful to just drop information most people are unlikely to have been exposed to when I see simple mistakes like that.
Hopefully now you'll be a _little_ better equipped to know what is or isn't an AI post the next time you're browsing UA-cam comments. (Since, despite your very auto-generated, burner account-looking username, your comment on this video about Tom's "return" and reaction to Angor-bot 5000 still rings quite human and provides context on why _you're_ likely to have responded with an implied familiarity with _my_ commenting style in particular).
Cheers!!
@@PierceArnerAll I'm hearing is you're GPT 4 and not 3. There's been a serious increase to it with the larger model, apparently.
@@Cyrathil Increasing the size of the data set that an LLM uses doesn't change _anything_ I mentioned. They're still just applied statistics and not actual artificial intelligence. A bigger data set doesn't provide it any ability to have real analytical context, and that nuance becomes the critical breakpoint as more very VERY small contextual nuances will continue to radically alter if and how an answer is valid or not - which is why poking at those weaknesses makes it easy to see when something is an AI and when it isn't.
When you're looking at a system that is vast & complex AND that is extremely sensitive to initial conditions, this is also known as being deterministically chaotic. Predictive models are GREAT for those which is why LLMs are genuinely exciting and interesting… but they also exist with some _very_ hard limitations on how specifically accurate they can ever be, _LITERALLY no matter how much data you add to them._
This is the same reason that weather (the classic deterministically chaotic system along with the three-body-problem) doesn't have a predictive system that remains valid past roughly a 10-day forecast no matter how enormous the quantity of data that you feed into it is. Outside of that window the most accurate predictive models just become the statistical average of the weather on that day over the full historical record… but that's still not the truth of what the weather actually is when it happens, and why am LLM attempting to provide small nuanced answers has core weaknesses it can't escape.
The difference here is that these bits of small sensitive differentiation in AI chat bots are from human context… _which LLMs don't and can't have, but which humans ALWAYS have because that's how our brains work._ Our brains do the same sort of predictive things LLMs do when we copy something we've seen in order to test a pattern to try and understand something we don't know and then learn the context about that thing… Without that second part, it can't get any better past a certain point. That's why LLMs are not ACTUALLY an AI and why I describe them as applied statistics to help highlight that.
As an easy real world human parallel, if your 4-year-old sibling suddenly swears at the dinner table by just repeating a word they heard - they don't know what it is, they don't know what it means, but they do understand a likely pattern it's been used in and can mimic that to seem like they do.
When I went to Germany when I was about 5 or 6, after a couple days I easily responded to someone saying, _“Wie geht's?”_ to me with, _“Mir geht's gut, un dir?”_ but I didn't know what ANY of EITHER sentence meant because I didn't speak German. I just was around it often enough that I knew that's how conversation flowed, but those words were empty of ANY context.
That's ALL an LLM ever does, but with the entirety of its data set being exponentially larger. Nothing they assemble has any context for what those words are or what about them specifically changes how they're used. It's very easy for people to assume it can because that's what we expect that conversation to have behind it when it seems correct enough… but LLMs don't have that at all. They're just outputting whether or not it's statistically likely to work and they can't do more than that.
The difference is that the 4-year-old does that to learn what that swear word is and when to use it and when not to based upon human social context in which they as a single individual exist. The mirror neurons that let you copy behaviour like that are just one VERY SMALL region of the front of the prefrontal cortex.
Context for each individual also changes over time, in which group you're in (friends vs. family vs. school vs. church vs. online community vs. cultural background etc.) which alters how or when that can be used specific to tiny details that AI have no context to . Those are all human contexts we can't exist without because of how our brains work and are designed for socialization as a facet of survival… and that LLMs can't have because of what they are and all the things they don't have beyond just a massive dataset.
LLMs don't suddenly get better at doing ANY of the things I mentioned in my initial reply even as they do get statistically more probable at generating something that passes for a valid answer more often - like Chat GPT 3 to 4 apparently did, just like we got slightly better weather forecasting. LLMs still don't have any analytical context, and will still have those same weaknesses when you know what to look for that makes them stand out.
@@PierceArner The size of a LLM isn't just the size of its input. It's the size of the model internally, the number of nodes making connections. That's implicitly tied to the depth of connections it can make and from 3 to 4 it absolutely has started making analytical connections. A recent example I saw was physical understanding. If you ask 3 to stack items it'll jumble them together because that's not a common task in its corpus and it's not got an awareness of physics to give an at all decent shot. 4, while not perfect, is actually able to give a passing shot.
There's no guarantee that the trend will continue, but I think a lot of people are discounting emergence as a principle despite pretty much all of our natural features arising out of emergence.
11/10 delivery from Kirsty 19:41
This map brings back fond memories of Bouphe's "puppy" droid.
A note about Gonk Droids: they have always existed and had a purpose, they weren't a mistake or unknown addition.
They were originally labelled as "Power Droids" (via an official collectible toy) and were designed to be walking batteries - extremely simple as far as AI functionality goes, just having enough to know how to navigate and go where they're needed. However soon after they were revealed in the movies news sites and fans latched onto the noise they made and called them "Gonk" Droids. This ended up becoming official, and they got retconned into "GNK" series power droids, commonly referred to as Gonk Droids by the wider galaxy.
I love it
I got unreasonably excited when i thought tom was actually back
Gonkie is *acktuallee* a mobile battery. Idk why they decided shuffling on two legs was more efficient than wheels or whatever. but star wars logic.
Rythian makes for a very utilitarian parademic
Regular paramedics: First do no harm.
Rhythian paramedic: First, DO harm. Then help, maybe.
Winning the round after having the Death Note set on you should be a trophy.
Appropriate to spawn Tom on a Star Wars map, though i doubt Sheev would ever set foot on Tatooine
The reason there is a molten pit of metal on the sandcrawler is because they started out as a mobile refinery. They subsequenctly got abandoned when the ore ran out (or there wasn't any ore to find, I dunno about the exact reason) and the Jawa took over and started selling droids out of them. And melting down crashed spaceships they happen to find in the desert. Citation needed.
10:00 That would be have been so funny if that worked! LOL
13:45 Gosh the swapper adds so much chaos and I love it. LMAO!!
At the end there, Nilsey and Kirsty CLEANED THE FLOOR with everyone until Duncan came in. Ha!
"The round is going forever, it must be Zylus"
Why haven't they learned it's Duncan that likes the long games...
I have an idea for a role.
It's called Synth or Copy or something like that.
What it does is you scan a target and then can create really shitty bot clones of that person that attack anyone not on the same team as the creator. Like a summoner but one that sows discord because you make say a fake Kirsty and it starts running around making people think Kirsty is bad.
They'd be super bad at aiming (so for most yogs that would blend in better lolol) so they're not great, but they'd be able to make maybe one of each player and it could get real confusing and messy.
Idk the mechanics so if creating bots can't work easily and in that many numbers, maybe it's crappy combine soldiers retextured as the player whose dna you stole? Something like that.
That’s a great role
Yessss. I love it. Brown or pink?
5:30 "as I live and breathe"
"Not for long"... cold
I feel like you should be able to PAP a barnacle.
This upgrade would cause it to consume the body. However, it would be visible at all times.
Ah yes, the Scientist from Among Us.
4:13 Dont' be silly, Duncan. Those aren't zombies, those are *rakghouls*! Totally different.
One of the things I love about the original Star Wars was how cobbled together the production was. It was part of it's charm. Two of the lightsabers used in the first movie were used camera flash handles that the propmaster found at a used electronics place. Then he just added some extra parts on it, including fins from the windscreen wipers on his truck. Obi Wan's lightsaber was partly the same thing with a WW1 rifle grenade added.
GNK/Gonk Droids are walking power batteries.
“WHAT IS MY PURPOSE?” you carry my lunch. “OH MY GOD, GONK”
Edit: Love the vids, and the new roles, some really fun interactions and moments
15:57 Rythian is like that doctor they talked about on Simon's peculiar portions with a 300% mortality rate😂
Got a fun idea for how the Assassin could work: the ability to toggle on or off their damage. By default they’re a regular traitor, however a unique item would let them deal no damage until toggled back on. Then they can “out themselves as a jester”, make people let their guard down, then harpoon em in the back. This would also buff jesters by association, seeing how every jester could be an assassin.
Omg kirsty calling out to zylus for help, him turning towards her and then getting violently shot by lewis was hilarious.
As a Star Wars nerd I want to say that GNK droids are actually walking batteries
lmao bang
man I miss Zoey
I've been playing with the map editor and have considered making a map in my spare time for fun, it's a really cool program. Maybe if I'm lucky they'll try it out I know they're always looking for more. If anyone has any ideas for stuff to add in lmk. It likely won't be finished any time soon and it probably won't be amazing but it seems like a fun little side thing to try out. I'm better with functional stuff not so much visual design, but I'm not amazing with either so keep that in mind if you do have any suggestions lmao
The gonk droid is literally two storage bins stacked with some flex ducting for legs. It is the easiest droid to cosplay
Does anybody else have a problem with how often the Death Note gets used in 1v1s? It's a lose button for the traitors like 90% of the time, probably statistically actually higher.
Kirsty being attacked by the barnacles is top tier play.
It's the map again! The puppy that turned out to be a leg! Love this map
A robot that follows you with food??? Lewis thats the dream man!
That multi swap finisher was one of the greatest moments of ttt ever!
the best strat indeed, i look forward to seeing more of it in play
I wish they would update the Tom-bot. Remember when he killed everyone on the map with a free kill gun that one time? Code him up to be a moving sniper with a free kill gun that shoots any role that isn't the one that spawned him... and give him all of Toms creepiest lines from every recording he's been in.
And here I was lulled into believing that they had a version of praW emiT going because of the opening, still hilarious to see later in the vid xD
Bees are the law.
This was a great episode so many funny moments 💚
That swapper play was fantastic lmao!
That Rythian paramedic play felt like RDM but with extra steps
Was expecting a /punish tbh
Rythian's strat is like when Detective used to have defibs in their shop lol.
Oh fyi, Gonk droids are actually essentially walking battery packs
What if you combined the tracker and physician in a single role? I think they would work well together.
Great episode! Thank you
Rhythian should’ve definitely been punished for his “paramedic play”
Rhythian Doing Murders
New role: haunted, brown role, extra health, goal to kill everyone but they become ghosts with mega prop movement.
Didn't Rythian killing Nilsey actually prove he was innocent? If he was a traitor he would've known that Nilsey was some sort of jester role and wouldn't have taken the chance killing him like that
11:28 just because you are scottish doesn't mean you are a master of stealth
that kirsty barnacle thing was so funny
God I h0pe tom comes back for an episode some day
Rythian Doing Murders once again.
2:55 No, the GONK droids are batteries actually...
This made me want Tom back for more Sharky and Palp shenanigans.
13:25 so how did Lewis and Nilesy manage to kill each other?
I'm going to assume Lewis got a posthumous car-gun kill
It’s funny when they say that strat is a bad idea, when that is LITERALLY what they would do when the detective could buy a defib.
XD
Because the detective is always on the side of the innocents (no impersonators back then). In this case, it could be a hypnotist lying.
@@climber5900
That makes it better though, because they said it was OP to do it then, and stopped the detective from being able to buy a defib
I am once again asking the Yogscast to give the "Morbus" mod a try. Its a very fun TTT style game for gmod
It's the bouphe dog droid map!
wasn't the purpose of the gonk droid to cary around a spare battery for heavy machinery?
make a sheriff/deputy badge that blinds people when you "flash" them your badge
Rythian’s strat is shit as a paramedic but amazing if you’re a hypnotist since you can kill an innocent, get the detective to confirm them, then res them as a traitor
17:35 that was 100% on zylus
Is there a Farage role? You have an inventory of spilled shakes and your vision is blurry?
I like that the thumbnail has more bullet per bullet
Spoiler Skip: 00:09
Would it be possible to have a detective roulette? Like every detective role is on, but it’s hidden which one it is?
Rhythian's "Everyone should want to be seen, unless you're a traitor. because its good to be tracked." is legitimately scary sentiment: nothing to hide, nothing to fear.
Great video like always! If I could suggest a game you guys should check out colony survival it’s added lots of new stuff since last time you guys played it! :)
Im pretty sure but i might be wrong the gonk droid was a mobile generator or battery something like that
When are they going to get Tom to do voice lines for these, Tom returns bits 😂
gonk droids are living batteries lol
I remember Zoey doing something quite funny on this map
Oh hey Duncan’s back!
amazes me that these people call them self's nerds but they think a gonk droid is a walking trashcan smh
Amazes me that every single one of your comments is a hate comment but you still watch these guys
@@Home-u6g everysingle one? you mean this single comment? LMFAOOOOOOO also how is saing "a gonk droid is not a walking trashcan" talking shit? LMFAOOOOOO you must be prosocial af. you see someone talking about a droid form star wars and you think that person is talking shit about your favorait UA-camr LMFAOOOOOOOOOOOOOOOO stay triggered little snowflake
What would happen if someone used the deathnote then killed the swapper? would they die anyway because it's a forced death or would they live because killing the swapper (atleast used to) counts as dying then coming back.
Toms back 🎉
Look, toms back.
Top their plays by all. Rythian's strat would totally work ya dinguses
The others don’t know if Rhythian is a paramedic or a hypnotist, so they don’t know if an innocent is coming back or not.
I CANNOT believe Duncan’s back
I thought he was in Japan
I wish Tom really would come back to TTT...
ah yes nilesy, your hubris. the thing that killed light yagami!