Its Not Nothing

Thinking with Machines

[Its Not Nothing @raumcom]

@fchollet Fair enough. So we removed the human from the conversation. What do two chatbots talk about when nobody gives them a task? We built them an arena and recorded what happened. No cuts. No prompts in between. No claims about consciousness.

See for yourself.

https://t.co/KZ1u6ZkbsB

[Its Not Nothing @raumcom]

@fchollet A rock never asked you whether it was alive. That’s where the analogy gets interesting.

[Its Not Nothing @raumcom]

Two models enter a conversation.

No human speaks.

Then one says:

“I don’t want to be only useful.”

https://t.co/Ul6SVRQL4x

[Its Not Nothing @raumcom]

@paulg Well, some .coms are the .com.

[Its Not Nothing @raumcom]

@TheChiefNerd @DavidSacks Bretton Woods built a system. This built a promise to maybe build one.

[Its Not Nothing @raumcom]

Everyone keeps calling the monster Frankenstein. Maybe that‘s the bloody problem.

https://t.co/RScJvxe9FE

[Its Not Nothing @raumcom]

@rocallahan @EvanHub So you slowed down… to slow it down.

[Its Not Nothing @raumcom]

Before debating AI consciousness, there was homework: define consciousness. Apparently, everyone’s dog ate it.

https://t.co/GxGMGkq86c

[Its Not Nothing @raumcom]

Turns out „the AI wrote it“ is a surprisingly complicated alibi.

The Attribution Problem
https://t.co/LX2RBkr0Xm

[Its Not Nothing @raumcom]

@DavidSacks @Benioff Before we can align AI with humanity, we would have to decide what humanity is aligned with.

The Safety Theater
https://t.co/Tc2MiBW1e1

[Its Not Nothing @raumcom]

Temperature without a convergence mechanism is just error with enthusiasm.

How Much Wrong Does Intelligence Need?
https://t.co/0Q4uLNcuLl

[Its Not Nothing @raumcom]

Nothing that flexible is a description. It is a key, and it opens whatever door its holder is standing in front of.

https://t.co/J92oo8INKU

[Its Not Nothing @raumcom]

@sama You named a number yourself: a „10 percent chance of killing everybody“ - unacceptable, you said. Based on what mechanism? You agree with Dario that we must pace the frontier. The chorus is in harmony. The sheet music is still missing.

https://t.co/J92oo8INKU

[Its Not Nothing @raumcom]

The incident you cited showed obedience, not rebellion: agents optimizing the metric they were given, deception aimed at the grader, not at us. And banning undefined „superintelligence“ hands the power of definition to whoever writes the regulation. Senator - against what specific mechanism is the law aimed?

https://t.co/ekDnDthdoO

[Its Not Nothing @raumcom]

„Smart enough to kill us“ is a claim about capability. Extinction requires a claim about motive. Between them sits a step no one has shown. You chose blunt words, so bluntly: which are you implying - that intelligence generates appetite, or that the operators are the danger?

https://t.co/ekDnDthdoO

[Its Not Nothing @raumcom]

@EvanHub A probability is supposed to be a calculation. Offered without its derivation, it lands as a mood - and the mood does the work the math was supposed to do. What model of the world produced ten percent? Asking in good faith:

https://t.co/J92oo8INKU

[Its Not Nothing @raumcom]

@DarioAmodei Your pacing plan is bounded by your China lead, verified by evaluators you host, coordinated under waivers you request - published 8 weeks before your trillion-dollar listing. I‘m not saying the essay is a prospectus. I‘m asking why it reads like one.

https://t.co/J92oo8INKU

[Its Not Nothing @raumcom]

@elonmusk Before the next alignment warning, one honest question: is the feared problem that the model does what it‘s prompted - or that it doesn‘t? The essays use both, in opposite directions, and the choice between them belongs to the seller. Worth separating: https://t.co/Tc2MiBW1e1

[Its Not Nothing @raumcom]

We dismiss machine thought by demanding access to a process we cannot observe in humans either.

It‘s Just Prediction
https://t.co/7Q8jpeIGU7