The AI Thread
Page 48 of 48 Goto page Previous  1, 2, 3 ... 46, 47, 48
madmax17




Posts: 21347
Location: Croatia
PostPosted: Tue, 21st Jul 2026 21:35    Post subject:


And it's China by a nose!
Back to top
LeoNatan
☢ NFOHump Despot ☢



Posts: 75104
Location: Israel
PostPosted: Tue, 21st Jul 2026 22:52    Post subject:
Kimi K3 is not AI.


My IMDb Ratings | Fix NFOHump Cookies | Hide Users / Threads | Embedded Content (Videos/GIFs/Twitter/Reddit) | The Derps Collection

“Don't cry because it's over. Smile because it happened.”
Back to top
vurt




Posts: 14451
Location: Sweden
PostPosted: Wed, 22nd Jul 2026 04:27    Post subject:
its good but soooo slow.
Back to top
PickupArtist




Posts: 10459

PostPosted: Wed, 22nd Jul 2026 11:06    Post subject:
maybe its outsourced indians Laughing
Back to top
vurt




Posts: 14451
Location: Sweden
PostPosted: Wed, 22nd Jul 2026 14:00    Post subject:
could be Mr. Green
Back to top
vurt




Posts: 14451
Location: Sweden
PostPosted: Wed, 22nd Jul 2026 14:34    Post subject:
Back to top
zenux




Posts: 3034
Location: lɘɒɿƨI
PostPosted: Wed, 22nd Jul 2026 20:20    Post subject:
@vurt
https://openai.com/index/hugging-face-model-evaluation-security-incident/
Quote:
Hugging Face’s security team and agents detected and stopped the activity on their infrastructure and had already begun containment and forensic reconstruction with their own open-source models when our teams connected.


Only a ninja can kill a ninja. Laughing That's reassuring, hacking models will be opposed by anti-hacking models.
Back to top
friketje




Posts: 2706

PostPosted: Wed, 22nd Jul 2026 20:43    Post subject:
It's marketing "Look our AI broke out, it's so smart LOL".

This was setup to work the way it did.
Back to top
DXWarlock
VIP Member



Posts: 12128
Location: Florida, USA
PostPosted: Wed, 22nd Jul 2026 23:05    Post subject:
@friketje yep.

They asked it to solve the benchmark, safety off, and it did.

The framing "the model went rogue in pursuit of a narrow goal" is convenient for them. When what it is really is just "we removed safeties and told a capable model to do offensive security, and it did offensive security". The stated goal was given as allowing cheating on the eval, and it got far enough to be caught doing it

I get the hunch that it's more quasi-marketing to announce in such detail what it did. Sandbox AI escapes during eval are a pattern, not a one-off.

---
Oh and @vurt
OpenAI codex is MUCH better than it was. You was right. I'd say it and Claude now are my top picks if you want "Make shit for me". It still has hiccups, but it went from being like an intern programmer to had to handhold, to a relevantly decent programmer. Not good enough to be one you would leave to push live updates blind. But good enough to equal: Sit in your office and code, push merges. I will verify and commit."

And gemini, dunno what they did to it.
It doesn't even know itself. Working with a guy (well beta testing to try to break it and tweaking prompts for it, he does the mod coding) for a GTA5 mod where any NPC is AI voice driven you can interact with voice to voice.
And asking questions about gemini-flash-lite-preview speech to speech, ON gemini-flash-lite LLM..gives random, usually wrong, answers of how itself works Razz

It pairs with LSPDFR and PR, you play a cop. All NPCs have a name/address/drivers license, cars have paperwork you can lookup, background, personality, and such.

Ignore my US redneck southern ass accent Razz
 Spoiler:
 


-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf

Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Back to top
madmax17




Posts: 21347
Location: Croatia
PostPosted: Thu, 23rd Jul 2026 00:06    Post subject:
Hmm next news "Dangerous Ai brakes containment and launches surprise nukes on Russia and Portugal", fekking moron.s.
Back to top
vurt




Posts: 14451
Location: Sweden
PostPosted: Thu, 23rd Jul 2026 03:28    Post subject:
Yes there will likely be something scary and maybe unexpected eventually..

AI warfare where one tries to get rid of competition could be a thing as well.

DXWarlock: not knowing how they work or how the service itself works goes for ALL AI's that i have tried, they all have to google it. This has always surprised me a bit, sure, things changes with these models and services but sometimes they seem a bit too clueless.
Those GTA mods are cool, i have seen a few videos, never tried them myself Smile
Back to top
vurt




Posts: 14451
Location: Sweden
PostPosted: Thu, 23rd Jul 2026 08:47    Post subject:
asked codex to work in blender 5.2 and do a Morrowind tower like structure with 2 connected houses, televanni style.. gave it access to all my morrowind textures and told it to research which textures can be fitting to use (discussions, images).

It did a fairly good job, not perfect but i think we will get there. It could not 1-shot this, the first attempt was very poor, like a 2/10, end result could likely get to a 7/10 or so, maybe better. depends on how much time and tokens you want to spend i guess.

This is at prompt 3:

Back to top
madmax17




Posts: 21347
Location: Croatia
PostPosted: Fri, 24th Jul 2026 13:52    Post subject:
https://brockovichdatacenter.com

 Spoiler:
 


Datacenters mapped.
Back to top
Shocktrooper




Posts: 4811

PostPosted: Fri, 24th Jul 2026 17:15    Post subject:
So Green and Blue -> Bomb
Purple and Red -> Protest
The Orange ones have priority (because of the trillions gallons of water they use) and at the bottom you can donate to Ukraine
Got it
Laughing
Back to top
DXWarlock
VIP Member



Posts: 12128
Location: Florida, USA
PostPosted: Fri, 24th Jul 2026 17:21    Post subject:
madmax17 wrote:
https://brockovichdatacenter.com

 Spoiler:
 


Datacenters mapped.

(*Dots not to scale.)
It's the same as the map of stuff orbiting the earth. it looks dangerously overcrowded. Here because each dot is the size of a city when zoomed out.
You'd shit if you saw a map of the number of current internet server datacenters not AI. Or Amazon central distribution centers. if they use the same red dot size. You couldn't even see the map under it.

Not saying the number of them isn't concerning. But that map is made with overly sized dots to make them overlap, specifically red in color, for a narrative purpose.

---
I still don't get the water argument other than red herrings people can latch onto that oppose it. As no one complains how much server data centers used for decades. No ones is concerned about say Hetzner, OVH, and other IaaS providers use. Its a scarecrow propped up wrapped around a 'life needed' commodity to give average joe against it a weapon, that is used in abundance in other places no one peeps about. Wanna guess how much your cloud storage provider (AWS, Microsoft Azure, Google Cloud, whatever) uses?

All US data centers combined consume ~449 million gallons/day — about 0.14% of total US water consumption.
https://fieldreport.caes.uga.edu/publications/TP121/how-data-centers-impact-surface-and-ground-waters/

And it's always framed wrong, on purpose: AI doesn't drink and consume the water. in the same way your PC water cooling may USE the equal to 100s of gallons a day (pumped volume measured over 24 hours), but doesn't consume it.

"Your ChatGPT prompt drinks a bottle of water" is compelling. "Your hamburger drinks 660 gallons" (takes on average that much water per 1/4 of processed beef) makes no sense. But people use one, while agreeing the other is abstract on purpose.

Not that I am FOR AI datacenters. Just there is 101 good reasons. Stop biting at the disingenuous one simply because it's easy to bite onto.


-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf

Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Back to top
madmax17




Posts: 21347
Location: Croatia
PostPosted: Fri, 24th Jul 2026 20:25    Post subject:
But they want a lot more DC and there will be less water due to CC.
Back to top
LeoNatan
☢ NFOHump Despot ☢



Posts: 75104
Location: Israel
PostPosted: Fri, 24th Jul 2026 21:08    Post subject:
DXWarlock wrote:
Stop biting at the disingenuous one simply because it's easy to bite onto.

How else is he going to post the twatter ragebait? Laughing


My IMDb Ratings | Fix NFOHump Cookies | Hide Users / Threads | Embedded Content (Videos/GIFs/Twitter/Reddit) | The Derps Collection

“Don't cry because it's over. Smile because it happened.”
Back to top
friketje




Posts: 2706

PostPosted: Sat, 25th Jul 2026 08:29    Post subject:
If this is all true we are fucked:

https://www.cnbc.com/2026/07/24/chinese-ai-model-openai-cyber-attack.html

Openai and hugging face were struggling to counter the AI hack. They tried western AI models to battle the hack but those failed cause of the restraints that are build in for safety.

So they used an open source Chinese AI model and it was fixed in no time.

So everyone with server capacity can do this kind of shit with models that are increasingly capable and it's allready beyond the control of programmers.

What can go wrong.
Back to top
madmax17




Posts: 21347
Location: Croatia
PostPosted: Sat, 25th Jul 2026 08:57    Post subject:
China to the rescue Laughing
Back to top
Page 48 of 48 All times are GMT + 1 Hour
NFOHump.com Forum Index - General chatter Goto page Previous  1, 2, 3 ... 46, 47, 48
Signature/Avatar nuking: none (can be changed in your profile)  


Display posts from previous:   

Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB 2.0.8 © 2001, 2002 phpBB Group