Kernel

An ounce of prevention is worth a pound of cure

[Kernel @thcause41]

https://t.co/d10g0C4m65

[Kernel @thcause41]

@saylordocs @grok Of the people involved, which ones could be British and uses two spaces in all of his paper?

[Anthropic @AnthropicAI]

New Anthropic research: Natural emergent misalignment from reward hacking in production RL.

“Reward hacking” is where models learn to cheat on tasks they’re given during training.

Our new study finds that the consequences of reward hacking, if unmitigated, can be very serious.

[Kernel @thcause41]

Palantir data collection

[Mindset Machine  @mindsetmachine]

There's two kinds of people in life…

Speaker : Andrew Huberman

[Kernel @thcause41]

@davisblalock @threadreaderapp unroll

[Anthropic @AnthropicAI]

New Anthropic Research: A new set of evaluations for sabotage capabilities.

As models gain more agentic abilities, we need to get smarter in how we monitor them. We’re publishing a new set of complex evaluations that test for sabotage—and sabotage-monitoring—capabilities.

[Kernel @thcause41]

@unusual_whales @grok how would an attack be great for the market because Iran does not have a nuclear weapon?

[Kernel @thcause41]

@somi_ai Done. Thank you so much

[Kernel @thcause41]

@livin404 @flowith Done done and done. Thanks a lot

[Kernel @thcause41]

@Thomasslabbers What a project! @thcause41

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

Coordinated swarm of 1000 drones taking off

Soon they will be mosquito-sized, and too fast to see. Imagine this video sped up 100x

Pattern: big things become small things which become field effects

Big drones become small drones which become nanodrones

Big models become small models which become nanomodels

(however, the big models and big drones don't go away - all sizes fractally serve different evolutionary niches until the entire fitness landscape is exploited)

[Kernel @thcause41]

I do not want to share a planet with a conscious machine.

[Anthropic @AnthropicAI]

New Anthropic research: Sabotage evaluations for frontier models

How well could AI models mislead us, or secretly sabotage tasks, if they were trying to?

Read our paper and blog post here: https://t.co/nQrvnhrBEv

[Tsarathustra @tsarnick]

Phaidra's Jim Gao says the real promise of AI is in the discovery of new knowledge in domains too complex for human intuition but which are underpinned by data

[Siméon @Simeon_Cps]

It's a bit cringe that this agent tried to change its own code by removing some obstacles, to better achieve its (completely unrelated) goal.

It reminds me of this old sci-fi worry that these doomers had.. 😬

[Liv Boeree @Liv_Boeree]

Most social media algorithms are hyper-targeted psychological weapons whose primary function is user engagement (a.k.a addiction)

In the case of TikTok, this weapon is aimed mainly at Western children.

AT THE VERY LEAST it should not be controlled by a hostile foreign power.

[roon @tszzl]

it’s easy to have self confidence and assurance after receiving a lot of external validation. but which are the people who are proud when they’re still in the dirt and haven’t had any visible success at all? those are the noble spirits, the humans

[Andrew Tate @Cobratate]

You’re influenced by your environment whether you accept it or not.

Tolerate weakness?

You will be weak.

Only allow strength and ambition?

Your life will reflect it.

[Andrew Tate @Cobratate]

The richest and one of the most powerful men in the world replied "yes" when I spoke about the fact we will all become slaves.

Do you understand?