Without any explanation as to why, the article references several developments in robotics that have seemingly zero relation to its primary claim as evidence in support of its primary claim. Color me unimpressed and unconvinced.
> We don’t have any obstacle left in mind that we don’t expect to get overcome in more than 6 months after efforts are invested to take it down.
Are there any institutions that share this belief? I don't think there was a Kanban board and all we were needing to do was to move the "obstacles" into "done" and then be ready for AGI.
Also lesswrong is generally one of the most absurd websites in existence, when it comes to users. Don't state specifically what's left to create AGI because it could be an "information hazard"? Give me a break.
You don't need an unaligned AGI for AI to cause harm to human society. You just need a bad actor to use the AI in a harmful way. Whether the harmful behaviour is a result of an aligned AGI or a human bad actor is irrelevant. It's like worrying if Trump or C3PO has access to a nuclear red button.
For example, I do believe it's reasonably likely that GPT-4 or GPT-5 could have super human hacking abilities. That's worrying because governments or even individuals with access to power like that could do huge amounts of damage. We don't need AGI for GPT-5 to be used in a destructive way.
And there's a lot of bad use cases like this. AI could be used to manipulate and scam humans. It could be used to create computer viruses. Potentially in the future it could even be used to create biological viruses. This is the risk we need to be focusing on ASAP.
I don't quite agree on the timelines here, but I do agree AGI is a risk. But AGI is only a risk if you already have AI models that are already advanced enough that they can cause harm.
No kidding. I worked for a while at a startup focused on creating robots that can do a fairly difficult, human involved industrial task. The software in question used existing robotic arms and attempted to automate the work.
The challenges associated with that project were some of the hairiest things I've seen in my career, and they still don't really have it working last I heard. If GPT-4 can even _help_ write a system like that in six months, I'll eat my hat.
Seriously, if by 8/21/23 you can demonstrate a robot perform a task that previously took a human to do, which is currently outside the mainstream scope of automation, using code that an LLM wrote, please email me a video to the address in my profile and I will record a video of myself eating my favorite hat. Subject to terms and conditions, etc.
I'm really not convinced we're close to AGI, at least in the traditional sense of it that LessWrong / the rationalists seem to be so worried about. I'm not convinced that our current approaches will actually lead to AGI in that sense.
However, I am worried about where AI is taking us. Because what I think I see happening is that we don't _need_ AI takeoff for this to change society for the worse.
Just look at the Microsoft OpenAI stuff for example. We don't need the AI to prevent its shutdown with grey goo and nanobots. The incentive structure around this whole thing means that even as stupid and sociopathic as Sydney is, they _won't_ shut it off. It doesn't matter that they _could_.
On the topic of alignment, the whole idea that _Microsoft_ of all companies thought that they could just feed GPT a good prompt and get the behavior they wanted is hilarious. We clearly have no idea how to get AIs to do what we want. It would be so funny that this was the approach of the absolute mind-boggling stupidity of it wasn't so horrifying.
I'm not worried about the AIs manufacturing a plague, or turning us all into paperclips. I'm worried that our systems of government and business will have no ability whatsoever to deal with the effects of these systems in a way that doesn't result in harm. And I don't even mean direct harm, from people losing their jobs (although that certainly seems like it's on the horizon). I just mean the same enshittification of everything that we've been experiencing for the last 10 or 20 years, at rapidly accelerating paces, further increasing the wealth gap and leading us towards a shitty, boring, unfulfilling dystopia.
It's not AI alignment I'm worried about, but human alignment. Looking back, it's miraculous that we somehow managed to dodge the bullet of nuclear MAD. Imagining the same thing playing out like that today is laughable.
I've been relatively positive about AI until the last few weeks, but watching the absolute shitshow with Microsoft has really made me cynical about the whole project.
