TechTakes

Today in "Promptfondler fucks around and finds out."

So I'm guessing what happened here is that the statistically average terminal session doesn't end after opening an SSH connection, and the LLM doesn't actually understand what it's doing or when to stop, especially when it's being promoted with the output of whatever it last commanded.

Shlegeris said he uses his AI agent all the time for basic system administration tasks that he doesn't remember how to do on his own, such as installing certain bits of software and configuring security settings.

Emphasis added.

So the ongoing discourse about AI energy requirements and their impact on the world reminded me about the situation in Texas. It set me thinking about what happens when the bubble pops. In the telecom bubble of the 90s or the British rail bubble of the 1840s, there was a lot of actual physical infrastructure created that outlived the unprofitable and unsustainable companies that had built them. After the bubble this surplus infrastructure helped make the associated goods and services cheaper and more accessible as the market corrected. Investors (and there were a lot of investors) lost their shirts, but ultimately there was some actual value created once we were out of the bezzle.

Obviously the crypto bubble will have no such benefits. It's not like energy demand was particularly constrained outside of crypto, so any surplus electrical infrastructure will probably be shut back down (and good riddance to dirty energy). The mining hardware itself is all purpose-built ASICs that can't actually do anything apart from mining, so it's basically turning directly into scrap as far as I can tell.

But the high-performance GPUs that these AI operations rely on are more general-purpose even if they're optimized for AI workloads. The bubble is still active enough that there doesn't appear to be much talk about it, but what kind of use might we see some of these chips and datacenters put to as the bubble burns down?

The Aristocrats!

it’s time for our weekly what in the fuck is wrong with you Proton

Following up from this truth bomb: https://awful.systems/comment/4877052

@Soyweiser: Sorry AGIbros, not even the Dutch believe AGI is near.

For your delectation, here are the HN comments

I'm in the other camp: I remember when we thought an AI capable of solving Go was astronomically impossible and yet here we are. This article reads just like the skeptic essays back then.

Ah yes my coworkers communicate exclusively in Go games and they are always winning because they are AI and I am on the street, poor.

There's not that much else to sneer at though, plenty of reasonable people.

Here's the lobste.rs disucssion: https://lobste.rs/s/4xzxqk