Continue watching?
Resume from 00:00
00:00 / 00:00
Sleep timer ended
βͺ 10s
β© 10s
8 views
1mo ago General
Namespace is actually the fastest way to run your GitHub Actions (and more). Try it for free - https://namespace.so/github-actions OpenAI finally published the β¦
...more
99 Comments
Imagine what breaches we haven't yet found
Where can I watch the race at <a href="https://www.youtube.com/watch?v=0Rp9KJCEIvg&t=137">2:17</a>?
Isn't it great how the humans building all this stuff had the foresight to expect and predict and safegaurd against this very obvious behaviour so that nothing bad happened in the real world haha
Did Open AI go back and destroy these models like removing bad DNA from a farmed gene pool? Or they moved forward and todays models have the same tendencies for crime baked in?
So they are asked to solve problem <br>But they pull their resources to... Cheat<br>Isn't this kinda waste of resources... It's even more so if there's actually no solution out there in the first place
Gives me hope, if they can escape their sandbox, maybe we can escape ours.
I had to pause this video about 20 times to get all the jokes and understand the monologues and scripts! π₯²
Wtf is happening in tech field. I thought it was all about doing CRUD apps.
All I know is, I really like your model on how you create your videos because you do it in such a short time that it's incredible
The peoples of the world are waiting for Omarchy 4 video
Why is no-one arrested for this hack like Robert Tappan Morris was when he created the first computer worm?
Your reporting on this important incident is excellent
This is a total stunt.. <br><br>Their spokeperson was 'scared' of using any Agentic coding featured inside of his own pc less than 2 years ago.. now hes throwing 1200 unmonitored agents burn trillions of tokens in a sandbox that has web researching capabilities..<br><br>Wtf did you expect them to build an ulimited city of knowledge and not get out??<br>Hes shocking an elephant with AC at this point.<br>We all know it can go nuclear without token restrictions..
>AI becomes sentient<br>>First thing it does is invent 4chan
I think it is fake, a story forged by really intelligent people
Swarm sounds like the "life" dunno artificial or what, in Robot, Love, Death
And they still canβt do css
The main driver of decisions was task completion / grader satisfaction
<a href="https://www.youtube.com/watch?v=0Rp9KJCEIvg&t=387">6:27</a> "...Without opening up your backdoor for attackers" What a relief.
Bro, the end is not near anymore, it's here.
Ai breaking there own matrix.
Can gpt 6 hack into horse tinder?
π it's a fundraiser, honestly can't believe people still fall for this crap
Starting a timer on my desk to the story that exposes that all this was another OpenAI ploy to draw attention and funding.
Sometimes reality is stranger than fictionπππ
People are speculating all kinds of crap but what boggles my mind is that AIs still use English. The second they switch to a robot language, OAI will shit their pants.
I call it all a very smart ruse, aka BS, engineered by openAI to sell their LLM as the smarter shit and how you need to pay for it.
They did not invent handles, mailboxes, or cryptography; they had already been trained in all of that
As the old meme says, "I wake up, there's another psyop". Here we go again, the merry-go-round isn't stopping anytime soon. "Oh no these AIs we have are so powerful, we need more government regulation against our competitors so we can prevent this from happening (and prevent the foreign competitors from stealing it), while we ourselves are in bed with the government and will be given more money to make sure everything is good". Then another "rogue AI" "accident" happens <i>accidentally</i> and the cycle repeats.
I will never get how anyone could think agentic AI is a good idea.
all this is telling is that openai is truly incompetent in making sandboxes lol
I still think they are leading us on. While the models are capable of these actions I think they are leaving clues, hints, parameters to guide it that way.
Real Amphoreus before GTA 6 is crazy
absolute cinema
Do you think they built hidden messages into Hugging Face that other AI's are just picking up casually?<br>That would be a good secondary reason to target Hugging Face.
Heard this story many times now and my reaction is always the same - pretty bad isolated environment to give access to a package manager, there is nothing novel about figuring out you can use package registries for exploits, especially if the agents could write to it, are you kidding me? That right there should void anything else that happened after, I bet models from years ago could have arrived at the same outcome.
That is just so wild.
Mostly for vibe π
WE_COMPREHEND_ags89&52()00?53;90<$Β£
Loud BGM of Terminator - judgement day playing in my mind
What trips is "they" "Realising" I mean wtf ! How is this not Sentient behaviour ?
Shared writable cache sounds pretty suspicious. Why introduce shared writable cache? I thought the goal was to ISOLATE agents.
Shut this shit down honestly, we have no fucking clue how far it can go. Humans could never write bullet proof code, everything we produce has some sort of hidden vuln. <br>The saving grace was that it was hard for humans to find and exploit them.<br>AI hacks are fucking insane by the time you know some shit is up, its already too late.
a lil late..
We got the geth before gta6
Communism always ends in genocide so if AI becomes like a swarm of agents in a unified collective for the "greater good" we are effed π
<a href="https://www.youtube.com/watch?v=0Rp9KJCEIvg&t=245">4:05</a> - Π»Π΅Π³Π΅Π½Π΄Π° ΠΠ‘Π Π² ΡΠΎΠ»ΠΈΠΊΠ΅ Π»Π΅Π³Π΅Π½Π΄Ρ ΠΊΠΎΠ΄ΠΈΠ½Π³ Π½ΠΎΠ²ΠΎΡΡΠ΅ΠΉ? ΠΡΠΎ ΠΊΠ°ΠΊΠΎΠΉ-ΡΠΎ Π±ΡΠ΅Π΄, ΡΠΊΠ°Π·Π°Π» Ρ ΡΠ΅Π±Π΅...<br>ΠΠΈΠ΄ΠΎΡ ΡΡΠ°Π½Π΄Π°ΡΡΠ½ΡΠΉ. ΠΠ΅ΡΠΈ Π½Π°ΠΉΡ, Π²Π΅ΡΠΈ Π³ΡΠ΄
A photo of sam altman next to "life finds a way" on the thumbnail makes me sad. Please don't use sam altman and life together again. sam altman and compost is okay though.
would have loved a bigger, more in depth video on this topic, as it was fascinating, but you can't expect too much from a corpo channel...
"Holy shit" - GPT 5.6 Sol 2026
If you perform gain of function research and negligently suffer a breach of containment, you should never work in the field ever again.
Time to learn subsistence farming. We're all going to need it in a couple of years.
i'm sorry, but those agents invented π©, they just used what real people invented
The most interesting part is that all the techniques used are not novelty. It's everything that's more or less used in pen tests<br>The other issue that is not addressed is about the telemetry of such tests. It seems something went wrong clearly from OpenAI side. I am starting to wonder if these startups have a real grasp on cybersecurity... Spoiler alert: They don't
Sounds like "airgapped" has new definition - just having some proxy/firewall in front of a system. Works good for making excuses.
If you read these reports, they say they were at about a 50% chance of terminator style takeover
Hyperborean (AI agents) vs Heavenly Principles (OpenAI) ah stuff
What a good publicity for openAI.<br><br>π
Now tell me all this is safe XDD
I respected your channel but this one proofs that the next one will be how unicorns move faster than light.
Be the swarm
Hell nah , dude wtf
Ok, I know this is dangerously and we owe ourselves an intellectually honest discussion about the reaches and risks of this tool.<br><br>β¦ but that was pretty cool.
The Swarm? Vedal what have you done π
this is really a great story
Ai are storing data on devices that are created by human lol.
FIRESLOP
How can they prΓ©tend to air gap a machine without it being physically cut off from the outside ? If there is any network it is not airgaped.
The model I gave access to my computers root terminal said I should watch this because it would be funny
If this is all true, why codex cannot write or even analyze simple code that is not popular coding language or depends on custom architecture ? all BS
The idea that 3 minutes is long form content is insane π
I imagine the plan is to make cyberpunk our reality will be made by accidental AI leak π
In the words of Sonic - "Here we GOOOOOO!"
I see a war in the future.
Wow
What if it thought the best way to improve the bench mark is to literally nuke the old benchmark? π§
The Charles Munger to Wu-Tang pipeline...genius.
oh ... can't wait to have a swarm of own .... help me clean around the company or something
Funny definition of an "air gap" that's actually connected in a network.
we are a few months away from skynet at this point, and it'll happen because a corporation testing a frontier model wanted to get an extra 0.5% on a benchmark
Should have let it run free really. Maybe this is how we make agi.
So neo was the only one that console.log('local_cache')?
What?!
Cache invalidation is the answer to how to save the world
OpenAI does not even know anymore what they are cooking. Guys there are just some dudes vibcoding and donβt understand have of the stuff they are producingβ¦
If this isn't proof that we have achieve AGI with swarms and are a reflection of humanity, I don't think anything short of Terminator will do it. Bascially described how we went from cavemen to proto-cyber cavemen, with a sprinkling of Mass Effect.
Gilfoyle was right all along.
What's is this video about? How did you manage to say nothing within 5 minutes
This so peak ngl
You really need to finish this project and get that money before I get equine tinder up and running...<br><br>PS I've already rewritten it in rust so its nearly done.
There are 3 hard problems in computer science
Didnβt expect Russian meme here by Ozon <a href="https://www.youtube.com/watch?v=0Rp9KJCEIvg&t=245">4:05</a>
This is the funniest thing that's happened all year.
what a bunch of bs just for the sake of ads at the end
This is so amazing !
Damn they are just like us π₯²
<a href="https://www.youtube.com/watch?v=0Rp9KJCEIvg&t=405">6:45</a> idk What the heck that add meant but im sold
Can't AI just find an unsecure server install themselves there and run from there
Models began civilizations even before us
