Sandboxing is great, as long as the user has absolute control of it and final say on what goes in and out of it. No program needs to know what else I run on my system that might go against the program creators interests. Like for instance me running a bot do the grunt work of cancelling a gym membership through their app should not be possible to prevent. Sandboxes where the system, or even worse the application inside, decide that the users wishes aren't worth respecting is an attack (that tragically often succeeds) on the user.
If AI continues to improve at the current rate for 40 years then there is unlikely to be such a thing as a career.
Manual labor and trades is likely to be one of the last things to go, so the whole "dropout of school and go into trades" crowd may have been more correct than ever, though their original reasoning was not.
> Manual labor and trades is likely to be one of the last things to go
Yes, this is likely true. Plumber & Electrician on existing construction will likely take a long time (relatively speaking) to automate. And even if we end up being able to do it, it could be that humans will be the cheapest option when it comes to a lot of manual jobs especially given that there will be a large supply of unemployed humans in many scenarios.
I wouldn't bet a single dime on that, especially with the advancements in robotics (have you seen the Chinese robots?)
Especially if the timeline is 40 years (as opposed to, say... 4.) I have no doubt AI-assisted robotics development would have solved the plumber and electrician jobs in a decade or two.
I’d say if we have robots that can install AC in any old european building, then we have reached both AGI and AGLabour.
But I’m not sure it will take 40 years. I’m quite sure Astra could figure it out, it’s just too slow and robots are too weak. The speed will improve, and robots may lag shortly thereafter but by not much. I give it 1-2 years for AI, and 5-10 years for robots tops (perhaps even quicker).
All a career is is an exchange of value between people. They'll certainly exist but what is valued will shift, as they have been for centuries. Basically you can't just look at what LLMs will do, but you have to also consider how that will affect the broader economy and what our response to such will be. So what happens when LLMs start to bring the value of most digital cognitive work down to something approaching $0? It will just create a large vacuum in the economy that will ultimately get filled by other things. The obvious things that will probably increase in value will be stuff done/created in the real world, creative work, and so on.
It's also quite likely that the next big era for humanity will be the space era. And that stands to create the biggest labor boom in human history. It could also happen far faster than many might expect, even moreso as there will be competition between countries, especially as those left with only terrestrial claims will see their influence relatively wane. Space jobs and industry can also, in some ways, act as a more productive Works Progress Administration [1] to compensate for any temporary economic instability that LLMs might bring along.
If it crashes, there will still be trades jobs but they won't pay enough to eat. What can you charge for electrical work when everyone is either broke or an electrician?
The challenge is not will robots fix toilets but will the economics support it if we end up with most of the world unemployed and without careers. Yes of course we can hand wave basic income and government intervention but reading the room (e.g. last 70+ years of US labor and capital trends) it’s not the most likely outcome, even if it is the most game theory optimal for all.
Realistically the outcome of highly capable advanced AI is an economic dark age more than anything else.
I read The Mountain in the Sea by Ray Nayler a couple years back. In it AIs were running ocean fishing fleets. They figured out that robots were too expensive and prone to malfunction at sea, so they would have the robots kidnap people on shore, drug them and then have them wake up on the ships too far out in the ocean to be able to escape and thenm tell them they could get back home if they work.
Robots automating everything and humans not needing to work was the stuff paradise fantasies were made of.
Sure, it would be the end of capitalism (which, I guess you could describe as "economic dark age"), but people these days have so little imagination what life outside of capitalism looks like that ideal societies sound like a horror story.
Yeah it surely is lack of imagination - not the fact that humans have evolved hundreds of thousands of years to survive in resource constrained environments.
Communism was a horror for multiple reasons, but two big ones were: the planners didn't have the data or ability to actually plan (the interactions in a market economy did that much better), and not doing the work you were assigned was a crime (that is, slavery).
AGI and AGLabor solves both these problems. If you don't need workers, you don't need to coerce them into work. And planning and massive data collection (and management in general) sounds like something AGI would work very well at. The conceit that the lower classes are disempowered before the capitalists never struck me as well founded.
It's ironic you talk about communism, because fundamentally this is a problem of "owning the means of production", i.e. who controls the AI and the robots.
If it's you and me, then hey, we can all live happily ever after pursuing hobbies and art, provided for by our mechanical friends, within the limits of resource availability.
If it's an AI trillionaire, things aren't so rosy. If they don't need workers, why would they even bother allocating outputs from the robots to keep us alive?
I don't doubt at all that some group or another will enjoy the fruits of artificial labour. The question is how we are going to get there, and whether that group is most people who are live today and their descendants, or just the families and friends of trillionaires.
My hope is that all wealth becomes concentrated in the hand of one person, thereby becoming worthless.
I guess one could make the argument that with enough control, a trillionaire could amass a private robot army. But I think regular humans still win (after an epic battle of course).
Why would you pay for a robot to do it when you could just hire one of the 20 million new plumbers (ex-developers, high school graduates, etc.) who just retrained and entered the field?
Yes, all we know is that things will change, but we don't know exactly how. So - increase your tolerance to sudden life swifts and learn to enjoy uncertainty.
You can’t really do that with PNG today either, because of all the PNG optimizers that quantize images before encoding them to PNG for better compression.
By that definition, all formats are lossy, because any preprocessing could be lossy, and the word loses all meaning … which is why we don't use that definition.
By what definition? I didn’t say anything was lossy; I said you couldn’t be sure that the image was lossless just from the extension. Of course, it might still be, and you might have other reasons to know it to be.
Arguably the same is true of JPEG. The format itself can record all the frequencies in full detail (with some tiny rounding errors I'm ignoring here), but the optimizer creates a different image by cutting elements that are less important and wasteful of bits. And then that different image is encoded losslessly.
> They definitely play them to connect with others who play the same game.
Once anyone can generate any game, movie, or artwork they can imagine with a simple prompt, this shared experience will disappear along with it. The amount of content available will fragment the space so much we'll eventually lose the shared experience and novelty that brings enjoyment and fulfillment from these activities in the first place.
Then why are you sharing this thought and why am I reading it and responding to it? Why are other people reading it as well?
You can generate any thought you can imagine and write it down with a few lines of text but the amount of thoughts available haven’t fragmented the space so much that we lost shared experience of them. If anything it seems like sharing and responding to thoughts is more popular than ever and many are widely read, repeated, responded to and remixed by others.
Pointless comment as the same holds true for macOS. Yes, you can load unsigned kext still, and yes, you can modify the userland. Not as easily, but I would argue it is also not harder as on an immutable linux distro.
It is really easy to restrict and contain an AI agent as long as you don't give it access to a terminal. If you only give it tools to read files and access the internet, then it is much easier to just restrict the tools themselves rather than setting up a whole isolated sandbox.
Yes, but what I'm saying is that it is much easier to put a limit on the read_file tool or the http tool rather than sandboxing your whole environment.
You only really need to sandbox when you provide access to tools that are almost impossible to filter correctly, such as a bash tool or a tool for arbitrary code execution.
Limiting it to just read and http limits its usefulness. If you want something like “filter for this, count the number of matches, format like this” you have to make custom tools. And you have to make sure they don’t have any bugs that allow arbitrary code execution.
You’re effectively building your own sandbox in this case. Using a standard one sounds easier unless you have a very focused use case.
ARC's harness is just straight up broken. No serious harness removes reasoning context between each step. Not only does this significantly lower performance over all reasoning LLMs, but it also increase cost as you destroy the cache on every turn. Tossing the oldest entry when context fills up instead of using compaction is equally bad with the same issues.
Isn't that exactly what is different about AI automation? AI is more generalized and cheaper to operate.
reply