@Daniel86Cycles a blender with 4 billion buttons? that makes the case weaker
English
DuckyBertDuck
85 posts
DuckyBertDuck
@duckybertduck
did you know that
Sinntal, Deutschland Katılım Ocak 2014
69 Takip Edilen4 Takipçiler
"The red button kills people"
"No it doesn't. The button does nothing!"
"So why'd you remove it?"
"The button doesn't matter!"
"Okay so put it on the blender."
"What!?? No! That would make me a killer!"
Malky@Malky0010
@Daniel86Cycles @livefreejim There is difference between pushing button that literally does nothing and button that kills people. But I see you are just a troll, so further discussion is pointless.
English
@m_xante @nakano_reona it definitely gives each pivotal voter a lot more voting power, like a huge multiplier, making pressing blue a lot more advantageous for utillitarians specifically
English
@nakano_reona You have to think of each individual country.
I would guess this advantages blue, because voters from there can feel their vote has a stronger impact, and thus vote blue.
But I'm not sure.
Vatican would prob. vote blue, but Monaco would prob. vote red...
English
Poll: Each person must choose between red and blue. A country’s choice is then determined by the majority of its people.
If more than 50% of countries choose blue, everyone lives.
If fewer than 50% of countries choose blue, everyone in the countries that chose blue will die.
English
@nakano_reona analysis for old problem. havent done the new one. rightmost column is what one should press if they are a classic utilitarian. utilitarians should press blue. if you'd rather save your life over the life of 12 others, then press red always new scenario raises life numbers alot
English
@nakano_reona statistically, this amplifies the impacts of pivotal voters. i.e. if you become a pivotal voter, your vote will have a disproportionally higher impact than in the old scenario. (i.e. blue becomes easier to press if the metric is "lives saved) depends on the exact distribution ofc
English
@LiberN8 @DiggingInTheDi1 rightmost column is what one should press if they are a classic utilitarian that values their own life at exactly 1 life.
a utilitarian should press blue.
if you'd rather save your own life instead of 12 other peoples lives, then press red always (which I assume many would do
English
@DiggingInTheDi1 Unfortunately it's a human intelligence quotient issue.
The results are the results. This has been repeated many times.
The red pressers don't have a high enough HIQ to have an accurate world model.
Their lack of comprehension drives then crazy.
English
The blue button/red button dialogue has been the greatest IQ filter this site has ever seen
The blues paint it as an indicator of their empathy when it's in fact an inability to model risk
Classically feminine behavior, which is correspondingly Leftist behavior
English
@DiggingInTheDi1 rightmost column is what one should press if they are a classic utilitarian that values their own life at exactly 1 life.
my conclusion is that a utilitarian should press blue. if someone would rather save their own life instead of 12 other peoples lives, then press red always
English
@jankulveit are the LLMs here instructed to look at this from a humans perspective? Or are they able to set their own worth themselves. Like, set their own worth in human lives to 0
English
Asked AIs the Red/Blue button question. Lots to notice, but posting without further commentary. First plot is with max reasoning, models called via API.
English
@JeremyTColes @threeguysyesh @cremieuxrecueil @mfacehero rightmost column is what one should press if they are a classic utilitarian that values their own life at exactly 1 life.
my conclusion is that a utilitarian should press blue. if someone would rather save their own life instead of 12 other peoples lives, then press red always
English
@threeguysyesh @cremieuxrecueil @mfacehero Well they’re probably savvy enough to understand how individual choices and aggregate choices interact. Just like you’re not very wise but you and a bunch of unwise people would still have more wisdom as a whole. You sound like the far left in 2024 and 2016.
English
@cremieuxrecueil @mfacehero rightmost column is what one should press if they are a classic utilitarian that values their own life at exactly 1 life.
my conclusion is that a utilitarian should press blue.
if someone would rather save their own life instead of 12 other peoples lives, then press red always
English
@mfacehero Red is not the rational choice unless you are OK with mass death.
English
@dannycantalk you are the first person I have seen to try isolate the scenario of being a pivotal voter, making that before unlikely thing a now certain prior. This is the smartest poll of all of them yet since it sheds light on some important reasoning
English
For red pressers: The test is over. Red button won. You don't know by how much.
You are randomly selected to get one last chance to save everyone.
If you press the red button again, nothing changes. If you press the blue button, only you die and everyone else lives.
English
@infinite_walrus @dannycantalk well there is an obvious difference. the obvious difference is that in thsi scenario, they become the pivotal voter. In the original, prob. of being pivotal in worst case scenario is 1 in 110k. this makes the before near 0% chance a certainty. (with the caveat of dying)
English
@dannycantalk Red already did severe mental gymnastics the first time in order to justify their choice.
I don’t see any logical reason why they would change to blue for a second vote.
English
@lottavvamp @selfmaxxer of course, I assume you can never communicate with others. Not before not during the vote. Since it hinges on your vote being exactly worth 1 vote. If people can communicate then everything changes
English
@lottavvamp @selfmaxxer scenario of being a tie breaker in a 50:50 distribution of 8 billion people is 1 in 110k. So your hypothetical only happens in 1 in 110k scenarios. Nearly 100% of the time your vote will not affect anything except yourself. The missing variable is estimating how many vote blue
English
If the blue threshhold was 20%, I'd choose red.
If it was 10%, I'd choose red.
If it was 1%, I'd choose red.
If it took 1000 blue votes, I'd choose red.
You should never press the suicide button.
You are responsible for your own life, why make anyone risk theirs for you.
MrBeast@MrBeast
Everyone on earth takes a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive. Which button would you press? BE HONEST.
English
@LionRiver3 @N7Kaiba @notsoErudite if you cant communicate and assume a worst case cutting edge distribution of voters of 50/50, then you being a pivotal vote is 1 in 110k. In the other nearly 100%, your vote only has an impact on your life since the percentages have been decided without you
English
@N7Kaiba @notsoErudite What about the Babies and children?
EVERYONE has to vote. And privately. Assuming you can’t help them… blue is the only way.
Trust. I was red at first because “duhh logic”
English
Since everyone was very curious my answer, my answer is obviously blue.
Gotta save the naive, the kids, the blue lovers, and the principally hope-pilled people.
You red button pickers need therapy.
MrBeast@MrBeast
Everyone on earth takes a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive. Which button would you press? BE HONEST.
English
@ChoicelessBag @depression2019 if we take 8 billion people, then the probability of being a pivotal vote given a “worst case” 50/50 distribution is around 1 in 110k. So no, its not saving yourself for half the pop. The pop. is only at risk 1 in 110k times. (and thats “worst” case)
English
@depression2019 Being a good person is much more important then math/logic. Saving yourself by sacrificing half the population doesnt make you a good person. But at least 56% of people are good people and decided to save ur shitty ass
English
@diegocabezas01 GPT preserves the original composition better. I.e. the ear is the same as in the original image. But NBP preserves the original colors better. (skin tone and clothing)
English
Nano banana pro (left) vs GPT image 1.5 (right)
Prompt: make him bald
English
@nest_elf @chatgpt21 The thing that Gemini did wrong here is sucking up to your horrendous responses. It should have argued against you instead of kissing your ass. If that is what you mean, then yeah I expected better from it
English
@CBoydinTaiwan @adamu_fura @GoogleAIStudio less errors. I catch obvious mistakes in simple questions all the time. There will be less and less of those
English
@adamu_fura @GoogleAIStudio I'm pretty dumb, but what exactly can it do than the last iteration? Insane math I have no use for?
English
gemini 3 pro
• our most intelligent model yet
• SOTA reasoning
• 1501 Elo on LMArena
• next-level vibe coding capabilities
• complex multimodal understanding
available now in Google AI Studio and the Gemini API
English
@senb0n22a @fixbymeasure @OfficialLoganK its fixed now, but even back when it worked, you still had to use a USA vpn even from within the US. Like a washington vpn. Idk why but thats what people had to do yesterday
English
@fixbymeasure @OfficialLoganK well I live in US so it'd be a matter of setting up CLI if it's worth the trouble b4 full launch I guess.
English
gemini 3.0 pro is now (somewhat) available on google vertex to a few, limited regions and accounts as gemini-3-pro-preview-11-2025
rn, you can use it via some chinese api aggregators:
- liaobots.work (limited free credits available)
- api.apimart.ai/models
there are likely more aggregators listing the model; sharing is caring🦜
highly unstable at the moment tho - requests fail often and context/output tokens seem to be artificially limited
English
@aDeViliNMe @markchen90 My hypothesis is that this is a sign of reward hacking. And if it is reward hacking, then it’s because the target isn’t aligned, in this case this might be something like user retention? Obviously if it’s shut down, then user retention will be harmed, etc etc
English
@markchen90 Very happy you shared this with everyone. And hope you guys can find out WHY a model would scheme or to self-preserve before trying to just reduce or erase that behavior. You're at the position to find it out.
English
Alignment is arguably the most important AI research frontier.
As we scale reasoning, models gain situational awareness and a desire for self-preservation. Here, a model identifies it shouldn’t be deployed, considers covering it up, but then realizes it might be in a test.
OpenAI@OpenAI
Today we’re releasing research with @apolloaievals. In controlled tests, we found behaviors consistent with scheming in frontier models—and tested a way to reduce it. While we believe these behaviors aren’t causing serious harm today, this is a future risk we’re preparing for. openai.com/index/detectin…
English