I enjoyed Paul Kedrosky describing the coming agent onslaught on businesses as 'a denial of service attack on modernity'.

To the point that there can be an aspect of what I want the LLM to do that I'm unclear on, but instead of thinking it through I can gesture at it and the LLM will think that part through. This is all sort of vague and a matter of degree, but I've noticed a real jump.

Something I've noticed with only the latest LLM versions: they understand ambiguities in my explanations that most humans would miss. I can be sloppier / less effortful in my writing than I would be if writing for humans, and still be correctly understood.

That...doesn't *sound* like what I believe, unless you mean that in some surprising way. I think there's likely been a communication failure of some sort. But in the end I suspect it may just come down to differing intuitions. Thanks for the interesting discussion!

lots of unknowns and uncertainty, eg around how to specify the relevant ways in which B has to be similar to A, but there are a lot of unknowns and uncertainty in every view on this issue, to the best of my knowledge. 3/3

B is likely to have the sorts of qualities that cause something to be consciousness'. There are certainly implementations of functional identity that seem less compelling, eg lookup tables (although I think intuitions about lookup tables are often misshapen). There are 2/

Who would you point me to that's making the claim that it is precisely 'being a simulation' that causes something to be / not be conscious? I take the argument from 'being a simulation' to typically be shorthand for 'A is conscious; B is functionally identical to A; therefore 1/

If you're paying to have the song I'm listened to *interrupted* so I can hear your attack ad you have 100% lost my vote

(postscript: I failed to connect back to the point I was trying to make. ...and so from my point of view, saying that something can't be conscious if it's a simulation is a category error. My consciousness can't depend on how someone outside me (who?) interprets my brain.)

/end

than others, and it may be that we can draw on patterns we find in neurons to create patterns of bits that are more likely to be conscious. But that isn't in any way determined, I don't think, by whether we interpret the bits as being a simulation of the neurons.

so its presence or absence isn't and can't be determined by whether we choose to interpret that in such a way that there's an isomorphism to (eg) some particular brain. It certainly seems like some patterns of bits or neurons are more likely to instantiate consciousness 4/

those bits as letters, or interpret them as part of an image, digits of pi, or whatever. There's nothing in the transistor state that determines which map to use. Consciousness, in my view, is a property of the transistors or the neurons or whatever the substrate is, and 3/

an *interpretation* under which those ink patterns are isomorphic to some part of the world, but that isomorphism isn't present in the ink or the paper. Maybe a better example: in some segment of computer memory, there's nothing but transistors. We can choose to interpret 2/

I haven't tried to quite articulate this before, so bear with me, but: no physical system is *intrinsically* a simulation. Calling something a simulation is an act of interpretation. To take the metaphor too literally, a map is just a bunch of ink on paper. There's 1/

Er, 2/2 Or rather 3/2 now

of water that modeled the British economy (npr.org/sections/planet-mone…). Humans chose to interpret it as a simulation, but physically it was just a bunch of water sloshing around.

Hmm. I guess my intuitions just land in a different place -- because 'simulation' feels to me like 'map', something that's a useful concept for humans, but not there in the world. Like I think of the MONIAC from ~1950, where someone created an analogue system from tanks 1/

that could be measured? And if so, what would an instrument for that be like? If I try to imagine such a thing, I immediately imagine something that's measuring the complexity and interconnectedness of the system, but that (it seems to me) is the functionalist view. 3/3

territory. In the world there are just systems. There's no external tag on systems that says 'map' or 'territory'. * I'm still really interested in this notion of sentience as a physical property. Am I misunderstanding, or does that mean it's the sort of thing 2/