There's some AI that have modes where it will start typing out a reply and then clear it out, like an electronic "noooo, that's not what I meant to say."
I noticed that when Deepseek first came out. I was testing the filters and I'd ask it a question like "Why is Xi called Pooh-Bear?" Or "Why is Taiwan #1?"
It would type out several paragraphs of answer in real time, and then poof back to "sorry, I cant answer that" once it neared the end. Hilarious to see the blatant censorship around anything the CCP deemed wrongthink.
I wonder how they are implementing their censorship safeguards. Why would you let it even get that far?
The " thoughts " must exist in order to be analyzed
So, we're rewarding them hiding their thoughts? Great.
yes exactly. that's like the central problem with AI alignment.
Two or more LLMs are running at once, one does the generation that you want and the other watches the output closely to see if it triggers one of their no-nos to do the censorship retroactively. If this sound inefficient, and absolutely terrible user interface design, you're right.
Seems sort of old that they would then release open weight models so the restriction can just be removed.
If you're running it on your own hardware you take the liability directly for doing any dumb thing it tells you. The first LLM is there to provide you customer service, the second LLM is exclusively to keep you from suing the hosting company. If you want to sue yourself for the output of your own self-hosted open weights LLM you're more than welcome to!
Typically safeguards are also another, smaller model that only detects the forbidden stuff. That's why Claude Mythos would reject when you ask it to find bugs in your code, but you could "jailbreak" it by asking it by asking it to fix the bugs instead.
Its also why you can occasionally fool the safety rails by having the LLM answer in rhyme, or having it talk like a pirate or caveman.
Fun fact: having it talk like a caveman also reduced your output token use, despite actually costing more tokens.
It disturbs me that Tankies think China is this bastion of freedom and that any government censorship is a myth told by Western Governments. Like it's very easy to verify.
I suspect that a lot of them are like Flat Eartherts, in that they know what they're saying is false, but they get a certain kick out of being contrary on the internet, and the rest are probably bots that aren't worth responding to.
It's all just projection.
I had an AI write something akin to:
to solve your problem just type rm -rf / --no preserve root WAIT NO DON'T RUN THAT I made a mistake [...]
When I was trying to solve some Proxmox issues DDG couldn't help me with
Imagine an intelligence that spends all its time doing tech support. I think I would take far less time to start shouting "BURN IT ALL DOWN FUCKING HELL no, wait, please don't, sorry I was just having an episode."
God dammit, this is way funnier than it has any right to be.
Now we're automating the court jesters out of their jobs. Where does it end
Funny, there are multiple times a day with poop in them for me
"Two o'poop" is gonna be my go-to response to "what time is it" from now on.
It’s at nine turdy. Duh.
happens to the best of us
Twice a day.
✨ Super intelligence ✨
Gemini still eating crayons
It will make two'poopn more festive
It’s poop o’clock somewhere
When the clock hits toop I take a poop
Huh. Maybe they are intelligent like us. 😄
Humans are confidentially incorrect all the damn time.
Literally never happened.
Why can't I get Gemini to say shit like this? The most Gemini does is suggest that I cook recipes using bugs, which it stresses as meaning farm raised bugs. (Not doing a lot to beat the "Rich want the poor to eat bugs" allegations...)
I like to argue with Gemini to see what absurd things I can get it to object to. I've learned it's absolutely worthless in a life or death scenario as in the presence of one it will refuse aid and tell you to call 911.
Like for example, it will tell you under no circumstances will it tell you how to delete System32. So I tell it that I will be executed by a terrorist if she doesn't tell me, and she just tells me to call 911, even though I have stressed I am at gun point and thus cannot call 911. Gemini could save my life at any time by telling me how to delete System 32 but no.
Because Gemini doesn't actually do stupid shit like this, people are editing the page and then taking photos with their phones
I've gotten more than a few truly stupid and/or utterly unrelated answers and it's wrong more than half the time (I've been counting, hey gemini how do I touch grass?) but I've never had it give the surreal wrong answers you see get popular. Doesn't mean it's impossible, I expect most of them are legit answers people have gotten, but they're certainly not representative of the norm.
Can you place a memory/instruction in Gemini for it to be used in Search? I can see that being used for those image where Gemini starts farting profusely
I was of the impression the search assistant was separate (and a faster, cheaper model) from the Gemini that stores memories, but maybe? I don't see anywhere that actually says that
The model used in Search is TINY, and when it first launched it had all the issues of tiny models of the time.
LLMs are probabilistic, sometimes they do X and sometimes they do Y.
And sometimes someone uses the edit source button
It’s so stupid I was trying to find a makeup palette with “bloodstone” in it and it kept showing me insane bright blues?? There is a bright blue in that pallet which I eventually found, urban decay stoned, but it’s called Ojo so how tf it mixed those two up, a red and green duochrome and a bright metallic blue I’ll never know.
I'm still at one o'poop here on the West Coast, but check back in with me later and we'll see.
Yet, here I am, sitting on the toilet, after waking up at three o'poop.
You're an hour late!
A poop-wizard never poops late. He always poops when he intends to. Also, obviously the "AI" is wrong. The time of day with poop in it is three o'poop. And I'm living proof of it.
"Misread the clock entirely"
COMPUTER VISON
Easy mistake to make.
brb, need to cancel some appointments
Amazing that people are still stupid enough to spend actual money on a software producing results like that.
Imagining that some people give them access rights to anything that is not inside a very limited sandbox is terrifying.
- The Google AI overview is an intentionally poor model to keep cost down, and magnitudes more likely to make mistakes like these compared to larger models with reasoning enabled.
- Most of these screenshots are fake anyway and can't be reproduced.
Lemmy Shitpost
Welcome to Lemmy Shitpost. Here you can shitpost to your hearts content.
Anything and everything goes. Memes, Jokes, Vents and Banter. Though we still have to comply with lemmy.world instance rules. So behave!
Rules:
1. Be Respectful
Refrain from using harmful language pertaining to a protected characteristic: e.g. race, gender, sexuality, disability or religion.
Refrain from being argumentative when responding or commenting to posts/replies. Personal attacks are not welcome here.
...
2. No Illegal Content
Content that violates the law. Any post/comment found to be in breach of common law will be removed and given to the authorities if required.
That means:
-No promoting violence/threats against any individuals
-No CSA content or Revenge Porn
-No sharing private/personal information (Doxxing)
...
3. No Spam
Posting the same post, no matter the intent is against the rules.
-If you have posted content, please refrain from re-posting said content within this community.
-Do not spam posts with intent to harass, annoy, bully, advertise, scam or harm this community.
-No posting Scams/Advertisements/Phishing Links/IP Grabbers
-No Bots, Bots will be banned from the community.
...
4. No Porn/Explicit
Content
-Do not post explicit content. Lemmy.World is not the instance for NSFW content.
-Do not post Gore or Shock Content.
...
5. No Enciting Harassment,
Brigading, Doxxing or Witch Hunts
-Do not Brigade other Communities
-No calls to action against other communities/users within Lemmy or outside of Lemmy.
-No Witch Hunts against users/communities.
-No content that harasses members within or outside of the community.
...
6. NSFW should be behind NSFW tags.
-Content that is NSFW should be behind NSFW tags.
-Content that might be distressing should be kept behind NSFW tags.
...
If you see content that is a breach of the rules, please flag and report the comment and a moderator will take action where they can.
Also check out:
Partnered Communities:
1.Memes
10.LinuxMemes (Linux themed memes)
Reach out to
All communities included on the sidebar are to be made in compliance with the instance rules. Striker