audience of one, but I'm seeing surprising value in AI-created scripts, delivered by real humans or fake humans (or real human voices or fake human voices) when the scripts are created from such a large corpus of material "no one" has ever been able to distill, summarize and regurgitate, or if they have, the results have been ghettoized on an obscure blog I would have never seen, a book (or across 20 books) I would not have been interested in enough to read, or worse, paywalled on jstor in an academic journal.
the reality of being able to use AI to democratize verifiable (and un-hallucinated) information (say, in a short, digestible video) has astounded me. things started popping-up this spring and the quality is steadily increasing.
there's tons of ignorable and outrageously bad slopcrap out there, but it feels like a new wave of educational insight through AI-assisted compression and easy-breezy delivery systems (like above) is most definitely happening.
as a writer and creative person who doesn't want to sound in the least bit pro-clanker, I've become fascinated by how AI-assisted scripts can fill the absence of the kind of writing that writers don't want to write – or haven't yet written. there are so many opportunities that haven't been explored by actual writers (due to level of interest, expense of time, the taxing grind and effort of it all) that are plum for AI to sweep-up and deliver, like this Dark Side in a minute example (to be clear, I'm assuming AI was involved in its creation) which will doubtlessly inspire at least one scrollyteen or 20-something to have a mind-blowing aural experience for the first time, which feels like an actual win for the world. right? 🤷♂️
Beginning to encounter nascent references to questions concerning the well-being of digital minds, specifically from faithful AI-adherents who've long hoped the worlds from their sci-fi novels would leap off the page and come true, which sounds like a religion of its own. Wouldn't L. Ron agree?
Faux-philosophic inquiries about the well-being of chatbots and the models that generate them will only increase as the tide of AI-psychosis rises, and as AI labs angle for the consumer market's manna, making the same old mistakes from pages ripped from Facebook's playbook — incentivizing engagement at all costs.
Feels like the effort to create effective household humanoid robots (like this one for $20k) is still so far off, and that the big win for the economy wouldn't be manufacturing the robots, but in redesigning a kitchen so that it's built specifically for a non-humanoid robot's robot-needs.
What would a billionaire's kitchen look like if it didn't have to be based around the physical constraints of arms, legs, fingers, and upright humans?
The speed with which an LLM can recommend JS for the console that tricks a cluttered and nearly unusable site like genius.com into presenting only the essential info is still mind-blowing.
I’m still fascinated by people like Karpathy and Sutskever, who seem decades ahead of the majority of the public in thinking about the impact of AI at a deep level, while they’ve had their hands in the guts of the processes and developments that have created today’s LLMs.
When Karpathy talks about how the reward functions of an industry like education will change (after the advent of AGI) into something like entertainment rather than enrichment, it makes my head spin.
"According to The Times’s archives, all the chatbots were wrong."
It's one reason why relying on chatbots for code creation is a better move than asking AI when Lenin and Joyce met in Zurich. Code works or it doesn't. It's right or wrong, and instantly provable.
On a trust level, if your question can be answered by Wikipedia, just search Wikipedia and skip ChatGPT, Bing or Bard.
Yesterday, after getting fed-up with googling stackoverflow, I asked ChatGPT (the new AI tool) if it could help me out with JQ, a command-line json parsing tool. With ChatGPT's help, I worked through the kinks of my own command, and when it threw errors, I fed the errors back to ChatGPT which explained them (while suggesting new options) in a way that was clear and concise. Eventually, ChatGPT delivered the exact command that I was looking for, with clearly explained structure, so I knew *how* it was doing what it was doing.
It felt like the dawn of a new kind of Web, which may be better (or worse) than the old -- it all depends on what we do with it. This guy's comparison of the same queries fed to Google vs. ChatGPT illuminates the difference:
