AI has already taken over #ai #attentioneconomy #bigtech #elonmusk #samaltman #joerogan

Video Description

Have you ever noticed that whenever you are fluent in a given subject you tend to notice that your AI companion seems to be… full of it? And even worse, people responsible for its behavior have started ‘tweaking it’ so that it sounds more human… which is insufferable. And even then, AI is too easy and too tempting not to use. Everyone who’s ever used language models like ChatGPT or Claude knows that it’s really easy to get used to using AI for writing, fact-checking, or even for learning new information. Even if it hallucinates or just straight up gaslights you, it’s still too attractive a tool not to use. Even as I write this I am constantly tempted to just write a bunch of notes and ask ChatGPT to write the script for this video for me. And, some would argue—why not? It is still your idea—you would simply be asking your assistant to do the boring part for you. The issue is that we, humans, really like leisure. In fact, we only default to actually thinking if we really have to. Whether we like it or not, our brains will always seek the easiest paths. And that’s not all that bad—that’s the mechanism that lead us to virtually every human invention that facilitates our lives. AC, cars, planes—you name it. And the worst thing is that it is literally programmed to act completely confident while gaslighting me and serving me its half-truths cocktail—why? Well, because the general user will be amazed by its performance—after all he or she doesn’t know any better. And so, more users will be amazed by how funny and person-like Grok is without realizing that it was programmed specifically to pander to the kind of people who like when their model talks like an edgy teenager. And companies will continue to do this sort of thing as a trick to make the user appeased and the shareholder happy. Because for companies that are legally obliged to their shareholders to make profit, truth is not of primary importance. Capital gain is. Across academic evaluations and real-world testing, hallucinations (meaning when your chatbot makes things up and lies) frequently exceed 15–20% — and can approach or exceed 50–90% in specific benchmarks.