Consider the political economy problems that might arise if (a) the U.S. government is financially dependent upon a thriving AI industry to finance ambitious new redistribution schemes and also (b) misaligned AI is an actual real-world problem but debated/very hard to detect. In other words, what if the “Yudkowsky school” is right and the only actual long-term solution to misalignment is some sort of an AI pause or ban, which USG structural dependence for new social welfare benefits would render politically impossible to implement.
If you are going to invoke existential risk or superintelligence as a politician with the level of seriousness and volume Bernie has, you had better be strategically and intellectually serious in your policy. This is transparently unserious, dodging the actual important questions anyone who had thought about the issue would have grappled with. Existential risk is window dressing for people like Bernie, and many others. Safetyists are sweet nerds they will bring in to add a scientific gloss to what amounts to good old fashioned schemes of property seizure and redistribution.
Here’s a fun one: there would be almost zero rankling about ai slop and ai use if there were clear and standard labeling involved so everyone knew what they were consuming. But scratch a little deeper and you realize AI cos don't want this because that destroys the value. What does that say? It’s a technology built on deception, theft, and displacement. i would buy the abundance argument if we locked down clear NUTRITION FACTS like labeling and categorizations and reorganized our various institutions accordingly, but again the goal is destruction not adaptation.
Publishing the reason chain would be nice but also whose gonna read all that?!?! 😩😩 and to my og point people who stop using it for writing. Right now id say 75% of ai writing use is to fool people into thinking youre smart, a good writer, or productive. Im just opposed that deception at a fundamental level
@matdryhurst Tl;dr all of these would bring ”ai” closer to actually being a tool and not just a general purpose spray and pray general purpose technology that requires its users to come up with use cases at the Point of interaction
@matdryhurst Another idea would be instead of adding new models that are ”smarter“ they could splinter them out to require more structured inputs to reduce hallucinations but then at some point this just becomes software or business process automation and theyd rather leave it open ended
“People waited for the next Dostoevsky installment like we wait for the next TV episode" is a point about distribution, not cognition. The medium still matters. Reading is an act of interpretation; television is primarily an act of reception. Please people read Neil Postman.
Chuck Grassley printing out the 12th draft of his 1,500 word oped red pen in hand, hunched over it in the wee hours the morning in a maddening search for le mot juste, all so he can file his draft by the editor’s deadline.