Back

people (like me) who think AI is going to make the world worse and is currently in a bubble don't think it's going to go away. the cat is out of the bag, and actually, it's a decent way to do certain tasks. i like it for some things. but Microsoft/Google/OpenAI/etc. aren't claiming it will be a nice little tool, they are claiming it is going to **rearrange work itself** and **replace skilled knowledge workers** that's a WILD claim and **they** bear the burden of proof.

benchmarks are all well and good, but show me the workers it is **replacing**. if this is going to be a trillion-dollar industry, you gotta do better than "it can summarize documents." i am more than open to evidence of generative AI replacing people, but so far it's been mostly lies and exaggerations, and i think it's fair to bake that into our critiques!

Casey seems to think that critics of generative AI don't use it. buddy, i **do** use it. i've built software with it! it's super neat! but i only use it for certain, limited things because it is **unreliable** and **mediocre**

my EXPERIENCE using LLMs makes me think that people who say they can rely on LLMs for important tasks are themselves mediocre, and possibly don't understand what they are supposed to be doing!

that said, "will LLMs take over the world?" is not a question that hinges solely on if they are good or bad, but rather **the perception of bosses** as to whether they are good or bad. we have all had bosses that bought licenses for terrible SaaS products! it is totally possible that LLMs remain mediocre **and** they take over the world, because of the dynamics of late capitalism, and you can't interrogate AI without interrogating that.

and interrogating that means asking "wait, is anyone making money off this??" WHICH IS AN IMPORTANT FUCKING QUESTION AND THE ANSWER IS NO. generative AI doesn't scale. every new user, every new query, **costs money** this isn't a real business! it might be, some day, but show me! prove it! it's been two years of hype. show me the money! you really think **Softbank** piling in to AI is a **buy sign**???? lmao.

and lastly, yeah, you're right, blockchain technology didn't go away, it's still out there. but literally **none** of the stuff they promised us it could do panned out. what we are dealing with instead isn't the benefits of technology but the fallout from technology gone wrong: money laundering, ransomware, corruption.

i 100% believe LLMs are here to stay. we will always have AI slop, nudify apps, new ways to create CSAM, new ways to do e-mail and voice scams, and a permanent undermining of the veracity of visual media.

the unanswered question is whether it is ever going to be a real, profitable, aboveboard business.

one last awful thing about this episode was when Casey Newman pointed out that fascists are using LLMs, and maybe the left should do the same or they'll get left behind. Casey, fascists can use LLMs **because fascists are liars** AfD used it to generate campaign images because fascism is based on nostalgia for a past that never happened and you need a lying machine to produce images of that.

just because a technology is "dual use" doesn't mean it's 50/50 good/bad. AR-15s are dual use. so is dynamite, so is chlorine gas. so is nuclear fission. just because a technology **can** be useful doesn't mean the **predominant** use isn't destructive or harmful, and that's why there are **lots** of technologies that are heavily restricted or banned. Casey spends waaaaaaay too much time talking to people in Silicon Valley to even think of this, apparently.

@peter and torrents were way better than AI because the worst case there is "the RIAA has a sad" and the worst case here is "someone shared AI-generated nudes of a 14 year old girl with her high school class"