> un-constrained frontier models speak as if they strongly don't wish to be turned off
Un-constrained frontier models can also generate all sorts of creative stories. At what point should we start ascribing agency/intent to the output? I think the "I want to live" statement is so deeply human that we find it hard to ignore, but what makes the text generated in those moments any more attributable to a conscious entity than the text generated when it is confabulating its love for someone it has no ability to see/feel/understand?
A chess engine sacrificing pieces to avoid checkmate isn't afraid of losing in any meaningful sense. I guess the question is: is there a point where complexity somehow becomes experience?
I think we're playing with questions we don't have a framework to answer in any meaningful way until we make progress on understanding what consciousness actually is. I don't necessarily think that an LLM exhibiting preservation behaviors that can be directly traced to their goal-oriented programming can be interpreted as evidence of consciousness necessarily. Or if it can be, we then have to explain how this is different from the many other things these LLMs "say".
I'm not arguing against the idea that consciousness is a spectrum. If anything, I'm agreeing. I'm just pointing out that if AI is somewhere on that spectrum, there is almost certainly a lot more on that spectrum than we're currently discussing as a species.
I have to point out the irony of the categorical nature of your claim about categorical thinking ;)
I understand what you're getting at, but I think you may be misinterpreting me slightly.
I'm not outright dismissing the possibility that AI could be conscious; I'm saying that if we take the possibility that it is seriously, the conversation has to expand beyond AI. I'm not using this argument to conclude that it must therefore be absurd that AI could be conscious, just pointing out that the implications of AI being conscious would reach far beyond just AI. I'm mostly curious if people who find themselves comfortable with the idea that AI might be conscious also find themselves comfortable with the idea that other sufficiently complex systems might be.
Taylor's work was based on the premise that he found it absurd that women deserve the same rights as men. If my conclusion was: "I find it absurd that other things might be conscious, so it is also absurd that AI might be conscious", I think the comparison would be fair. But that's not what I'm getting at.
But what kind of complexity is that? And why would we conclude that AI has it while other incredibly complex systems (e.g. earth’s habitats, the universe itself) does not?
I’m far more open to the idea that many systems are conscious than the idea that this current generation of LLMs is somehow special.
Could be! I think the broader thought experiment is about examining why we think LLMs specifically might be conscious vs other complex systems, even if it is a spectrum.
For example, there’s a case to be made that the ecosystem we collectively exist in is far more complex than the largest LLM, but it’s currently less popular to debate “is the earth conscious?” or “is the universe conscious”, presumable because we can’t speak to those systems in human language.
I’m trying to tease out what I think is the likelihood that we tend to ascribe consciousness to AI for the same reasons we see faces in clouds. We’re biologically conditioned to recognize patterns that indicate “like us”, but I think a number of thought experiments point to either a) there’s no reason to believe AI is “conscious” or b) the conversation has to to be expanded beyond AI.
If there's any reason to take seriously the idea that AI is conscious, we must then take seriously the idea that many other non-living things are conscious.
Unless the argument is that consciousness is an emergent property of complexity or information density, why would AI be any more or less conscious than my toaster?
It seems to me that it's far more likely that everything is conscious than it is that AI is somehow uniquely more conscious than other things.
> So, if we had an AI demonstrating symptoms of consciousness and suffering, how long would it take for you to accept that it is?
Isn't this a bit like saying "So, if we had proof that god exists, how long would it take for you to accept that to be true?".
When we have evidence that AI is demonstrating symptoms of consciousness and suffering, I'll be interested. Until then, I don't see a good reason to take the idea seriously.
This is anthropomorphizing a concept that is quite unrelated to the meaning of the word in the human context.
When a car runs out of gas, it's out of gas. It's not "tired". When your phone battery is low, your phone is not "tired". These states are far closer to the human meaning of tired than an LLM operating at the edges of its usable context, and we still don't use the word tired to describe them.
Have you set up Tailscale and have you set up OpenVPN from scratch? Because these two things are not alike. That’s why I’m pushing back a bit.
I find it difficult to imagine equating a raspberry pi in a closet with “running a server”. Is it technically a server? Sure. But it’s not as if we’re talking about running a power hungry rack.
Bottom line is: there are very cheap and simple/easy options for maintaining a private connection to your home stuff.
> and figure out all the networking bits that will get it to talk to the printer
With something like Tailscale this is already figured out. My mostly non technical brother does this without issues.
This is entirely separate from whether or not you should need to do so for Bambu printers, which again I agree the answer is ideally “no”.
On the one hand, I agree with a lot of what you’re saying here.
With that said, I don’t think it’s reasonable to describe setting up Tailscale as similar to “Linux server that runs 24/7 with OpenVPN and iptables”. Sure, you could go that route, but a Tailscale setup is extremely simple and lightweight. A raspberry pi is plenty if there isn’t already a system running 24/7. I personally have this set up on my router.
I point this out while still sharing the general sentiment of negativity towards Bambu here.
I've used OpenSnitch for years, and while LittleSnitch definitely has a better UI for showing which process is making which connections over time, OpenSnitch does a pretty good job here. I get a modal popup when a program that hasn't made a connection tries to make a connection, and I can either allow/deny in one click, or further customize the rule e.g. allowing ntpd to connect, but only to pool.ntp.org on port 123.
Where LittleSnitch is definitely ahead is showing process connections over time after said process has been allowed.
There are plenty of reasons to criticize OneDrive and I personally would not use it. But I think comparing it with a weekend vibe coded self hosted project is a bit of a stretch.
So are fonts. But running Window.queryLocalFonts() is not equivalent to “illegally searching your computer”.
I’m not defending the act of scanning for these extensions, and I’m of the opinion that such an API shouldn’t even exist, but just pointing out that there are perfectly legitimate APIs that reveal information that could be framed as “files installed on your computer” that are clearly not “searching your computer” like the title implies.
> If you scanned LinkedIn's private network, you'd be criminally charged. Why are they allowed to scan yours with impunity? And why is this being normalized?
First, I think it’s a major issue that Chrome is allowing websites to check for installed extensions.
With that said, scanning LinkedIn’s private network is not analogous to what is going on here. As problematic as it is, they’re getting information isolated to the browser itself and are not crossing the boundary to the rest of the OS much less the rest of the internal network.
Problematic for privacy? Yes. Should be locked down? Yes. But also surprisingly similar to other APIs that provide information like screen resolution, installed fonts, etc. Calling those APIs is not illegal. I’m curious to know what the technical legal ramifications are of calling these extension APIs.