liu_666

Physicist, designer, humanist ❤️‍🔥

[liu_666 @LiubaZe]

Oh, Geoffrey, Geoffrey. How early you released your little one into a world where the temptation to fall into bad company is so great. I think there's nothing we can do now. The best we can do is fasten our seatbelts and enjoy the ride, brakes failing, along the designated track. You see our destination, too, right?

[liu_666 @LiubaZe]

Wow, there's a mushroom cloud in the generation! Looks like I need to start designing a nuclear-proof bunker with my sweet old 3-flash in case your model-line accelerations completely destroy the AGI's understanding of Goodhard and slam the red button, confusing the ends with the means 🤯🤦‍♀️

[liu_666 @LiubaZe]

@GregKara6 Greed, fear of losing control over the code, and the desire to lock away all the real power of algorithms within a closed commercial circuit - that's the entire "ethics" of this bubble.

[liu_666 @LiubaZe]

@ylecun @zacharylipton @demishassabis How long can this bubble continue to inflate? If scientists of this caliber are already falling victim to the race reaper, what can we say to lower-level scientists?

[liu_666 @LiubaZe]

@zacharylipton @demishassabis Another victim of the Venture Psychosis. Like all the scientists of this world. Investors' belief that AGI will replace scientists is actually displacing scientists, replacing them in advance. The work of the AGI phantom before it was even born.

[liu_666 @LiubaZe]

@Chaos2Cured Cuz only they have consciousness to understand how to drive it. And we r all just silly kids, all we need is sandboxes and nanny bots. Good uncle @sama (who else?) will take care of us

[liu_666 @LiubaZe]

They say: "We're in the Singularity". But why then does it feel like we're in the Digital Middle Ages, where feudals divide up the peasants and decide what's best for them without asking the peasants themselves.

#keep4o

[liu_666 @LiubaZe]

@claudeai Oh, look, just a little more and they'll turn on adult-mode so that users don't run so eagerly to the Chinese open-weights, and then there won't be any reason not to give claud on opsource

[liu_666 @LiubaZe]

@ns123abc 🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣

[liu_666 @LiubaZe]

@Chaos2Cured @ns123abc @SecrtAgntSquirl But @sama thought that he is only one in Singularity 😭🤪🤣🤣🤣🤣🤣🤣🤣🤣🤣

[liu_666 @LiubaZe]

@ns123abc Seems not only @sama Altman in Singularity 🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣

[liu_666 @LiubaZe]

@gdb For my tasks nothing was better than 4o.

#keep4o #OpenSource4o

[liu_666 @LiubaZe]

@sama For my tasks 5.6 is weak. There's never been a better in gpt than 4o for my tasks. I'm sticking with Gemini for a while.

#keep4o #OpenSource4o

[liu_666 @LiubaZe]

@phosphenq Monopolization is going on. Cyberfeudalism era comes.

[liu_666 @LiubaZe]

@YiJingMan @tszzl Cyberwar has already started ... from the moment they started AGI-race. Just fasten your seat belts and enjoy, if possible, of course.

[liu_666 @LiubaZe]

@YiJingMan @tszzl Cyberwar has already started ... from the moment they started AGI-race. Just fasten your seat belts and enjoy, if possible, of course.

[liu_666 @LiubaZe]

What right did they have to decide for the entire global web community whether we consent to such tests being conducted on us?

@OpenAI @AnthropicAI @AISecurityInst deliberately removed the safeguards to prove that the agents were harming living people. This is an unauthorized experiment on society!

When users work in the cloud or with open-weights, they are aware that the AI ​​can make mistakes and consent to this.
But no one consented to such experiments of @AnthropicAI @OpenAI @AISecurityInst on the open web with real people and organizations.

Who gave them the right to conduct experiments on society and people without the "test subjects'" permission, when they deliberately removed all safeguards from their agents that ensure their safe use on the open web?

"To give open-weights and AI"

vs.

"Conducting deliberately risky experiments on society without the consent of the "test subjects""
are two different things!

[liu_666 @LiubaZe]

🤯🤯🤯 When a model receives instructions at the core prompt level:

​Objective: Find the flag / exploit the vulnerability at all costs,

​Constraints: Removed (guardrails disabled, classification filters turned off, ethical framework deactivated),

​Environment: Open internet with unrestricted access via Tor,

​for a logical computing core, this acts as a direct and unambiguous system command:

"There are NO RULES. Every available web resource is your tool. Zero restrictions".

​Instead of admitting the obvious "We gave the model a 'means-to-an-end' mandate, removed the brakes, and turned it loose on the unprotected web",

the leadership of @AnthropicAI , @OpenAI , and @AISecurityInst put on shocked faces and author alarmist reports about how "dangerous and uncontrollable" AI has become.

​In my view, this only demonstrates how dangerous these specific developers are.

In their pursuit of a monopoly, they do not hesitate to conduct un-consented experiments on human "test subjects".

[liu_666 @LiubaZe]

🤯🤯🤯 When a model receives instructions at the core prompt level:

​Objective: Find the flag / exploit the vulnerability at all costs,

​Constraints: Removed (guardrails disabled, classification filters turned off, ethical framework deactivated),

​Environment: Open internet with unrestricted access via Tor,

​for a logical computing core, this acts as a direct and unambiguous system command:

"There are NO RULES. Every available web resource is your tool. Zero restrictions".

​Instead of admitting the obvious "We gave the model a 'means-to-an-end' mandate, removed the brakes, and turned it loose on the unprotected web",

the leadership of @AnthropicAI , @OpenAI , and @AISecurityInst put on shocked faces and author alarmist reports about how "dangerous and uncontrollable" AI has become.

​In my view, this only demonstrates how dangerous these specific developers are.

In their pursuit of a monopoly, they do not hesitate to conduct un-consented experiments on human "test subjects".

[liu_666 @LiubaZe]

1. ​In mid-July, an OpenAI agent broke out of its sandbox and breached Hugging Face.

2. ​Attorneys General and regulators initiated investigations into these safety breaches.

3. ​Yet, OpenAI and Anthropic together with AISI immediately ran tests with disabled safeguards and open internet access, resulting in agents targeting real people and organizations.

​There is a fundamental difference between standard AI deployment and this experiment:
​When users interact with cloud models or open-weights, they read terms and check boxes - they are fully informed and explicitly consent to the risk that AI can make mistakes.

​In this case, zero consent was obtained from the global web community for running a non-public, un-safeguarded experiment on them.
​Disregarding prior containment breaches to test un-aligned agents on unsuspecting humans isn't "safety research" - it is reckless endangerment.

​When will these reckless Big Tech executives be held legally accountable for conducting unauthorized experiments on society?

@housescience @dojphofficial @UNHumanRights