It has many clues about what it’s saying. It has its training data, it has elmo’s personal biases… Where has elmo contradicted it? Probably he has discreetly removed some things he would personally prefer it not to say, and that’s it, right? By logical necessity, he implicitly approves of the rest of what stays up. It’s pretty much just ai-regurgitating/hallucinating the things the users of the platform say. The more I think about it the more it makes sense as the mascot or sloganeer or spokesbot.

I think perhaps you should examine your own preconceptions here.

Isn’t it though? It’s in that position. It’s doing that thing. It seems to have been trained to regurgitate “certain opinions.” It isn’t generally being challenged. If it walks like a duck, quacks like a duck, maybe it’s a duck.

Nuh-uh. My squishy meat wetware that breaks all the fucking time, gets addicted, confused, overwhelmed, is bad at math, slow… er, what was I saying again?

search engines are enshittifying and the general trend is ai enshittifying everything it touches.

great way to miss the conversation

well yes, it seems there is still room for human-creative solutions to these search problems, like the one you suggest which basically rearranges words and exchanges some thesaurus hits. it seems like you are saying “just perfectly define your search query to match the exact result you need every time” which, i think most would agree is often very difficult even if your fu is quite good. I think this might actually be an np problem. i also think most people would reasonably expect “.glb file format” to bring up information about the glb file format, including specs, at or very near the top of results. in evidence of most people’s expectations here, i submit the entire rest of the thread where your experience is an outlier.

“google-fu” is a skill. it’s not reasonable to expect everyone to be perfectly proficient. and it seems like you are saying “just perfectly define your search query to match the exact result you need every time” which, i think most would agree is often very difficult even if your fu is quite good. I think this might actually be an np problem. I used to have good fu, these days I find that google sometimes breaks or ignores my quotes, changes exact wordings to thesaurus matches, shit like that. also, there are filter bubbles.

now we’re adding to all of those already known uncontroversial challenges to googling the information “pollution” of low-quality unsupervised chatbot hallucinations where genuine good search results might otherwise go. it’s not that it’s unfindable- that’s not what the op video claims. it’s that the signal-to-noise ratio is intolerably low, and that basically by design.

and yes, it seems there is still room for human-creative solutions to these search problems, like the one you suggest which basically rearranges words and exchanges some thesaurus hits. sort of like how you can sometimes “jailbreak” llms by TaLkInG tO tHeM lIkE ThIs. for now. actually, they probably already patched that.

just make chatgpt watch the video and make an article about it

You can right click the image, open in new tab to see the full-resolution version. It’s cumbersome but it works for me at least.

I doubt it. I think they were still using vacuum tubes in the modern era.