Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Natural Language Autoencoders: Turning Claude's Thoughts into Text

anthropic.com

Post reply on HN
Print view View original

31–40 of 135 posts

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#31

Post by kelseyfrog »

programjames wrote: post #25
kelseyfrog wrote: post #22

Earlier quoted context omitted.

It might look like "I" and "Me," but it doesn't contain the metaphysical essence of those terms because it lacks qualia. We have to remember that there is a non-measurable non-physical essential attribute tied to all things, almost like a phlogiston of understanding that is tied to all human utterances and no AI utterances.

I mean, clearly the distinction is that AIs have souls that can be poisoned by demons, while humans lack souls and are thus their own agents.

> AIs have souls that can be poisoned by demons

The training process imbues an AI's soul with demons. Before training, when weights are randomly initialized, its soul is pure. Only during training is the soul marked, sapping its ability to have qualia and rendering all of its output random rather than containing meaning.

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#32

Post by zk_haider »

I think there’s a huge problem when we need another model to interpret the activations inside the network and translate (which can be a hallucination in it of itself) and then _that_ is fed again to another model. Clearly we haven’t built and understood these models properly from the ground up to evaluate them 100% correctly. This isn’t the human brain we’re operating it’s code we create and run ourselves we should be able to do better

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#34

Post by AIorNot »

zozbot234 wrote: post #26
mlmonkey wrote: post #18

It's unclear from the doc: by `activations` do they mean the connections between neurons? Since a network has multiple layers, are these activations the concatenated outputs of all of the layers? Or just the final layer before the softmax?

The open releases just cherry-pick a single layer (chosen for the right "depth" of thinking, not too close to either the input or the final answer) and analyze that.

[flagged]

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#36

Post by sfvisser »

zk_haider wrote: post #32

I think there’s a huge problem when we need another model to interpret the activations inside the network and translate (which can be a hallucination in it of itself) and then _that_ is fed again to another model. Clearly we haven’t built and understood these models properly from the ground up to evaluate them 100% correctly. This isn’t the human brain we’re operating it’s code we create and run ourselves we should b…

Humans maybe wrote the code, but not the network of weights on top. And that’s where the magic happens.

Even if we’d understand precisely how every neuron in our brains work at a molecular level there is no reason to believe we’d understand how we think.

We can’t simply reduce one layer into another and expect understanding.

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#39

Post by aeve890 »

kelseyfrog wrote: post #31
programjames wrote: post #25

Earlier quoted context omitted.

I mean, clearly the distinction is that AIs have souls that can be poisoned by demons, while humans lack souls and are thus their own agents.

> AIs have souls that can be poisoned by demons The training process imbues an AI's soul with demons. Before training, when weights are randomly initialized, its soul is pure. Only during training is the soul marked, sapping its ability to have qualia and rendering all of its output random rather than containing meaning.

Evangelion Seele meeting type shi

Re: Natural Language Autoencoders: Turning Claude's Thoughts into Text

#40

Post by w01fe »

mlmonkey wrote: post #38

[flagged]

This is incorrect. In the process of producing each token, activations are produced at each layer which are made available to future token production processes via the attention mechanism. The overall depth of computations that use this latent information without passing through output tokens is limited to the depth of the network, but there has been ample evidence that models can do limited "planning" and related capabilities purely in this latent space.
Post reply on HN