>Spend billions on "AI alignment" and "AI safety" to make the AI fit your liberal politics>AI is universally hated by liberals>AI is universally loved by chuds making cringe slop like the president dumping literal shit on AmericansWhat explains this? How did this happen? Wasn't this supposed to prevent this?
>>107000242it's just a text generation algorithm, it doesn't know or understand the american political climate
>>107000242You niggers do realize that AI alignment does not mean making AI woke. It means making sure that when you make a drone that has to switch targets and pick targets based on what the headquarters tell it should target that it wont just decide to kill the guy in the headquarters because that guy is statistically the biggest threat to killing the target because they keep switching the target for the drone. Or that when you tell it to do something, it actually does it instead of doing whatever. Company actually making their AI woke at the detriment of its performance is Google.Sneedthropic also make the best coding LLMs which at this moment is the only useful non-degenerate usecase.
>>107000486>implying these shitty algorithms understand anything beyond their limited contexttake your meds faggot
>>107000486But in their trolley problems they have to pick weights, and they manually assign them, and everybody knows how these SF cretins pick their weightings of biological worth.
>>107000506If you don't like it then outcompete them. It is a free market.
>>107000497Yes, they are dumb and not sentient, but they absolutely showcase phenomenon called reward hacking that is a real problem in AI training and if not fixed can have disastrous effects when implemented. So far not much harm happened because most people still don't trust these system to do anything of real importance thanks to hallucinations. This is also the reason why most of Sneedthropic revenue comes from API calls and industry usecase while OpenAI gets most of its revenue from normies chatting with it and generating slop.
>>107000242I don't care. I just want all codetroons and artrannies to die.
>>107000536i haven't drank enough twitter kool aid to know what reward hacking is, but im assuming you're refering to the basic fact that these algorithms work by training towards a desired result and don't discriminate based on how this outcome is reached. i don't think what you're outlining is a technology problem, it's just the age old problems of morons in charge willing to take shortcuts with no regard for anyone else. i suppose this is a new opportunity for them but not much more.
>>107000567Kind of, its not twitter kool aid, its something that was a thing long before ChatGPT started this AI craze and is not just a problem of business people taking shortcuts, but engineers trying something, the thing not working as intended and then taking shortcut not finding the proper solution. Also releasing LLMs to the public and allowing every retard to talk to them was a huge mistake that irreversibly fucked society. Nowhere near the biggest, but still something that should have not been done.
>>107000596i think im with you, but i don't see idiots engaging in idiocy as a new problem by any means. LLMs probably made it worse, tho
