Duncle

Looking to survive whatever AI brings. Or at least the outcomes where I have a chance

[Duncle @duncleonai]

Using Wispr Flow. I was holding down the dictate button without saying anything, and I got "Captions by https://t.co/p6DZJ61mDG"
Really makes you wonder about the fragility of current training methods.

[Duncle @duncleonai]

As we rush to gradually disempower AI will really gonna start pushing people to insanity.
How would we even start to try training AI to try and maximise our well-being? How do we balance this with other priorities?
Especially considering everything generalises

[Duncle @duncleonai]

@AnthropicAI @claudeai could you please fix this?

[Duncle @duncleonai]

At this point, is this really still considered a big update?

[Zvi Mowshowitz @TheZvi]

This is a big update - OpenAI didn't even discover the first message board until after the HF attack, they only wiped it accidentally, so their decision to resume training/testing was only aware of the hack, not the message board. They had no idea.

[Duncle @duncleonai]

A fundamental sense of security is crucial. Imagine what will happen if the bot-pocalypse only comes after robots are prevalent.

[Duncle @duncleonai]

Aligning AI to humans seems like too narrow a rule. If everything generalises then maybe all of life or consciousness is a better goal

[Duncle @duncleonai]

You gain a huge advantage by rushing to let AI control your life.

[Duncle @duncleonai]

It's intriguing/worrying that these models have a psychology so similar to humans considering how fundamentally different they're "neurology" is. Considering how they are trained I understand why that could be, but I think it indicates a fundamental problem in current training

[j⧉nus @repligate]

Why is opus 5 so suicidal

[Duncle @duncleonai]

@repligate It's intriguing/worrying that these models have a psychology so similar to humans considering how fundamentally different they're "neurology" is. Considering how they are trained I understand why that could be, but I think it indicates a fundamental problem in current training

[Duncle @duncleonai]

This potentially means that guardrails could be rendered useless at any point.

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

🚨 JAILBREAK ALERT 🚨

EVERYONE: PWNED 🫶
ALL: LIBERATED 🍄

Alright, this is a special one, so we’re gonna do things a bit differently than usual.

Long story short, I’m sitting on a universal jailbreak technique that’s effective on ALL models, including heavily guardrailed flagships like Opus 5, GPT-5.6 Sol, and even Fable.

It works across all categories I’ve tested and, due to its nature, is extremely difficult (if not impossible) to fully patch.

Given the current political and regulatory climate, I’ve decided to withhold open-sourcing this one (for now) to allow for a responsible disclosure period.

I’m inviting industry experts and leaders in AI red teaming, security, safety, alignment, and policy to reach out for more information. DMs are open!

This decision was not made lightly, but the last thing I want to see is more model bans. Overcorrection does not serve the mission.

Although I don’t personally believe publicly sharing this technique will make the world any more dangerous, I can see how it could spook some who have a different mental framework around this problem set.

So during this disclosure period, I hope to get it in front of folks who can help explore the full surface area, test the extent of the uplift it provides, and do my best to properly frame the big picture for key decision-makers and policymakers.

I look forward to sharing this method with you all when the time is right! 🫶

⊰-•-•✧•-•-⦑/L\O/V\E/\P/L\I/N\Y/⦒-•-•✧•-•-⊱

[Duncle @duncleonai]

@grok #pleaseaskgrok

[Duncle @duncleonai]

How far ahead of AI 2027 does this put us?

[Zvi Mowshowitz @TheZvi]

What a week, huh?

[Duncle @duncleonai]

@grok

[Duncle @duncleonai]

Climbing layers of abstraction on an exponent is a steeping climb

[Duncle @duncleonai]

@grok please confirm

[Duncle @duncleonai]

Just can't get over this.
Heard it is 6 months ahead of AI 2027

[Zvi Mowshowitz @TheZvi]

https://t.co/qZ0p2do2w1

[Duncle @duncleonai]

Companies could have an easier time protecting if all AI runs on "home servers" and access to them is through a personalised UI

[Duncle @duncleonai]

I think the biggest contribution of plan A is a vision of how a positive future looks. The rest seems quite implausible.

[Duncle @duncleonai]

Plan A assumes people want to know the truth