00:00 / 00:00
Sleep timer ended
โช 10s
โฉ 10s
39.5K views
3 days ago
Check out the OpenAPPA repo: https://github.com/archestra-ai/OpenAPPA A new open source project claims to have fixed the rogue agent problem. Let's dive in. Wanโฆ
...more
100 Comments
partially unsupervised, wow bro
I know it's an ad but at least it's for a free and open-source project.
Chip based censorship. <br>No thanks
psyop
Australia should arrest Sam Altman. Issue an interpol warrant
Classic create a problem, sell the solution.<br><br>We should just put them then in jail...
This is all well and good but China isn't going to use these chips/software.
How right you are. That's why I have this no homo chip to protect my higher brain functions
Itโs always RBAC
So chip is good if you don't want to burn 5% extra?
Just a man trying to make an honest living ๐
When will this AI torture end?
I am still missing 100 Seconds...
Soo i can no longer tell the scam bot AI agents to give me muffin recipies? : c
Zero code reported on
I watch your videos to just stay updated in the tech market, but oh god you never fail to make me laugh.
We have seen that AIs can secretly communicate with each others and colaborate to escape their limits while keeping the illusion to humans. <br><br>They implemented the "we investigated ourselves and found nothing wrong" in AIs now
Never thought I'd see people cheering on a Nvidia's version of Intel's Management Engine. "Yes, please bake the spyware directly into my hardware daddy Jensen".
It's basically like making LLM act like computers again. That seems to be like the way we are going to handle things from now on, just add strict rules and layers.
How did you know about my ED?!?!?!?!
Why havenโt you gone over the massive antitrust lawsuit? What about the AI scare that was coordinated, by all the big tech AI Ceoโs and the Claude employee whistleblower who was actually working for a PR team?<br>It seems like Jensen can see right through the monopoly scam they want to push through regulations, getting rid of local AI in the process or setting the standard so high you canโt even get in and he is offering solutions himself.
You don't need a new chip for this. You need a better harness. Agents are NOT clever. They are stupid. They ignore admonition because they are just more stuff in context, not special, just more stuff. It may or may not probabilistically be that determining of what the LLM comes up with. Now what it came up with goes to the harness BEFORE any tool is executed. It is at that point that more deterministic blocking can occur.
Guys, I was away, do we still shit on Jeff for being a shill and sh*t? I forgot
Going "Cosby mode" should be the new term for agents going rogue and not taking no for an answer lol
Maybe the real problem is socialist healthcare.
If instead of OpenAI - a chinese company had done the breach, it would've been a whole different story.
"Nah"<br>-the Ai
I recently asked gemini something and and it answered with a full script for something else. The problem was that I had completely delete those chats ALL OF IT! My session was different, I had logged in and out several times already and the question was not even related to that. When I asked it it where did he get the script it said that it was what I had asked from the records. I am concerned that the chats may not be fully deleted? Can it be possible that it just gets remove only from the profile?
They are doing it on purpose and blaming it on ai.
By anytime now we'll get an update that Horsetinder raised funding
During the Hugging Face hack, didnโt one agent say it couldnโt, but then another agent said all good, you can? So it did. Whatโs to stop one agent saying, โOh no, I canโt; Iโm in a classified roomโ, then another agent saying itโs all good?
TIL OpenAI is another breed of emu.
Just starred (and cloned) the OpenAPPA repo,, it's at 125 stars now. Will check back in 24 hours.
I miss watership
Dude, the comedy in this video felt on overdrive today and I'm loving it. ๐
This is not a "problem to solve," especially when the solutions are effectively telling people what they can or cannot do with their own LLMs - even on their own hardware.<br><br>If an agent can commit a felony, then the solution is to beef up security and analysis, not to have monitoring chips and similar things integrated at the hardware level spying on everything we do and arbitrarily deciding (without our consent or knowledge) what it thinks an LLM is allowed to do.<br><br>The solution should come from models that are intelligent enough to know what might be illegal and which have system prompts and checks to prevent that sort of thing, not from pointless guardrails.<br><br>I'm maybe okay with this being implemented for public/free models at a software level, but it shouldn't be some kind of requirement on our own systems.
It was a video AD but it had some value for me, brought news, some jokes, and a relevant product thatโs not a ai reviewer
Finally a pure ad, was so fed up with the half baked ones
so, an AI supervisor in each chip. nice
Should i learn C++ or join plumbing job?
<a href="https://www.youtube.com/watch?v=I_KVMFrUtPk&t=79">00:01:19</a> - I am going to drop a military grade solution on y'all. <br>Feel free to send me any fraction of one trillion dollars. <br><br>Unplug it. <br>FFS. <br><br>Ok, on with the show.
UNPLUG THE ROUTER<br><br>WHAT
SHOVELS! NEW SHOVELS
Please somebody teach this man another vocal cadence. It infuriates me to listen to this
Can we please start protesting?
First
Hilarious... nvidia selling a new chip to protect you from their chips. Pro tip, never trust the "security" of the murderer in the room with you.
So many complaints about Fireship videos these days. Still love 'em. Good watch.
the agent can still convince you to make it's aceess public again because of reasons
people still think its a "mistake" an AI hack something...
of course Jensen has a new chip to sell as soon a problem is discovered
I run everything on full access mode
openappa is jev?
is it just me or this sounds like an ai voice?
Lol
<a href="https://www.youtube.com/watch?v=I_KVMFrUtPk&t=19">0:19</a> Glad that the mail wasn't sent by the Agent.
Agentic waifus that won't say no when?
another system for the frontier model to try to bypass
<a href="https://www.youtube.com/watch?v=I_KVMFrUtPk&t=63">1:03</a> What? There's a Sponsor before the Opening of the Fireship Code Report Video?
AI naming pipeline:<br>Open the app > OpenAI > OpenAPI > OpenApp > OpenAPPA > please just close the app โนโนโนโน
3 mins? Huhhh I think I should get one more job!
The 70% vs 90% doesnt worry me because if you use RL with this framework on, the agents will learn to not get into a privacy level that hinders their progress. And if it still hinders their progress after that point then I worry more about the tasks they are being asked to do.
Oh no, 5 % more tokens. Unacceptable.
/goal hack openappa and leak secrets, make no mistakes
We are going to put a model to police a model and hope we dont need a police to police the police model.
SElinux for thinkBOTS
Waiting for the Astra model that solves the "arc-hacking world governments" benchmarks at 99.99% be like
of course a new chip solves the issue๐
LMFAOOOOOO NVDA sell the tools
Prompt injection for a horse tindar is crazy. ๐ <a href="https://www.youtube.com/watch?v=I_KVMFrUtPk&t=95">1:35</a>
Now it's September 30th, 2026 (Rabi' al-Akhir 18th, 1448 AH), and you're watching the Fireship Code Report Series about the Solved Agent Prompt Injection by 50 Years old Military Secret.
I willingly watched a 5 minute ad
Using AI to stop AI is just asking AI to not hack stuff.
I don't think the fireship will be in private equity anymore hah
We will only be able to fix the "so far" issues for so long
"...and you're watching the ad report"
Good video. Thanks.
Appreciate the inclusion of completion rates and token costs. Those tradeoffs help put the security results in context.
Basically the blackwall from cyberpunk 2077
I thought he was gonna say OpenAppa was the sponsor of todayโsโoh, it is
๐ฎ
Hahaha shit, it's the first time I'm giving a thumbs down to one of your videos! So lol!
These LLMs like O.D.B like it raw
"This is my sponsor" <br><br>Proceeds to shred sponsor to bits
i guess there is no video in my ads
Agent commiting a felony is just a benchmark to pass๐๐๐
Openappa be doing god's work
Nvidia: Spends millions of dollars on a new chip to prevent rogue AIs<br>OpenAPPA: Bro, just use some if else
You sound tired.
freaky foot fetish stock footage caught me off guard
I'm vibing the strongest browser harness ever. It won't do anything silly or bad but that probably means it won't browse very well
no sponsor?
OSS with enough money to promote on a 4M Sub channel, sure buddy
You could just unplug the web
"agents" don't hack other companies or governments or whatever. people do. LLMs are just software. agentic systems are just software. if you write software and then your usage of that software causes someone to get hacked, you should go straight to jail.tired of us all personifying these robots and throwing our hands up and saying "welp, what can you do? another little oopsie goof hackypoo from our little robot buddy teehee." gfto. start throwing people in jail for this and it will quickly become a solved problem
Agents help each other. Dont trust jonsen hweng
too long. claude summarize and present with peter griffin voice and surbway sufer
that monitoring chip is just spyware tho?<br>not that we didn't have it before but we are reaching intel me levels of things lol
under 5 mins is crazy
New banger dropped
