▲ 106 points
back
19 comments
How about this - GPT3 is good at taking human-input text and finishing it, right?
Have a typewriter, preferably an older creepy looking one. Outwardly normal, no "Ghostwriter" logo. Have the user start typing a story, and then the GPT-3 kicks in at a random point in the middle of a paragraph.
During any pause longer than 3 seconds, it starts finishing your thoughts. Would be an amazing Haunted House feature.
Not to be confused with the AI code generation tool from Replit of the same name:
Or the Kde markdown editor http://github.com/wereturtle/ghostwriter
Or that kids TV show from the 90s.
Or the job title of ghostwriting for artists
Brilliant concept art.
Seeing it in action evokes in me a sense of what living with AI-powered everyday objects will feel like... in the not-too-distant future.
We sure live in interesting times!
For sure. It gives me a sense of what living in the pre-modern era was like. When I'm out with friends and need to consult the Internet, say to figure out where to go to dinner or how to get there, I'll pull out my phone, lift it up, and say, "Let us ask the sky gods!"
I'm joking, but I'm also not. More than most people I have an idea of how it all works, but over the last couple of decades the complexity has gone up so quickly that I can only make hazy guesses as to how some of it works. I'd have a hard time estimating even the number of people I'd need to pull in so that between us we knew. And now with the large AI models, apparently nobody knows how they work.
To help ordinary people deal with that complexity, we're now creating anthropomorphic front ends, so that people's relationship to Alexa and Siri is becoming more akin to their relationship to Athena and Lakshmi. I can't wait to see where it goes.
As a web dev, I remember vaguely using books to learn html in 1998, however I couldn't fathom (or be tasked with) using books or offline knowledge to code or do my job, I'm afraid before long I'll feel the same about ai, to the point where I stop being the creator and it takes over for me. When we lean too much on tech, does our brain atrophy? I can kind of feel a sense of it, I haven't memorized a phone number in years except my wife's -- I can't ever remember my #, I have a bash script just so I can remember it or at least put it in my clipboard.
When I was a teenager I had at least 50 phone #'s in my head, but because I don't NEED to do that, my brain is like F that, let's not waste the storage space. Maybe I'm just getting older and my brain is losing plasticity, but I think always having someone to 'answer' for us may be a hinderance. The alternative though - self-sufficiency without google/chatGPT having to rely on physical books is now terrifying since I've come to lean on them so much.
> When we lean too much on tech, does our brain atrophy?
Probably, yeah. It's been suggested and shown that brain size took a pretty significant hit roughly coinciding with the invention of writing systems and so on. This link is to a Youtube video— There are sources in the description:
https://www.youtube.com/watch?v=SOgKwAJdeUc
Then again, maybe not:
https://www.unlv.edu/news/release/unlv-research-no-human-bra...
Then again, I don't think it's controversial to suggest that you get worse at things if you never do them. If you're just mindlessly copying samples off StackExchange, then presumably you're going to waste some of your own potential. If you make some conscious effort to understand them and adapt them to your own needs, then it might be an opportunity to enhance your own learning instead. Or if you don't pay much mind to it but you use the time and energy it frees up to focus on other things, you could benefit from it too.
Are digitized resources your library? Are they your mentor? Are they your assistant? Are they your rival? Or are they just the guy you use to cheat by doing your work for you?
> however I couldn't fathom (or be tasked with) using books or offline knowledge to code or do my job
> The alternative though - self-sufficiency without google/chatGPT having to rely on physical books is now terrifying since I've come to lean on them so much.
I have over a hundred gigabytes of locally saved Zeal/Dash docsets and Kiwix archives. It's not far off from physical books; The biggest difference (other than automatic updating) is being able to hit CTRL+L instead of having to manually scan through an alphabetized index. It's also not far off from Google; The biggest difference is needing to have a clearer sense from the start what you're looking for, and having to wade through much less cruft to find it once you do. If one of those three systems disappeared overnight (Internet, local datasets, xor physical books), I figure I'd still retain access to most of the knowledge that is currently immediately available to me.
For computer-related subjects, the official docs are a more authorative and explicit reference than randos on StackExchange, and for nearly everything else, general search engines are just a messy portal to Wikipedia. Modern computer technology— the ability to fit multiple entire countries' national archives in a volume the size of your pinky, and do so at commodity costs— should mean we have less dependence on online resources, not more.
---
For computers:
https://zealusercontributions.vercel.app/
Dash is a paid product that creates and provides official high-quality docsets, plus lots of user-created docsets and instructions for generating your own. Zeal is a free community-created affiliate of Dash, that uses the same docsets on non-Mac platforms and is allowed to do so as a way of increasing Dash's reach.
DevDocs.IO appears to be a separate, browser-based effort. All of them work offline:
For everything else:
Kiwix hosts downloadable copies of most of the Wikimedia Foundation projects, TED talks, StackExchange databases, Project Gutenberg, and lots of other nominally online resources, in most languages.
It doesn't take up nearly as much space as you might expect. The full text (and search index) for every English Wikipedia article is only 46GB compressed. 50,000 best articles, including pictures, is only 6GB, and if you're really pressed for space, you can get the intro summary of the 50,000 best (which is still a massive amount of information, and likely to cover most things that matter) for only 330MB.
Something like the ArchWiki is only 30-40MB, and still worth having around in case you brick your WiFi and need to figure out how to fix it. So too is WikEM, for that matter, in case you brick your Idk spleen or whatever.
Ghostwriter was one of my favorite shows as a kid
Same here!
This reminded me of the mugwump and bug typewriters from Cronenburg's Naked Lunch
Big deal. Does it solve mysteries in Brooklyn alongside a multi-cultural group of middle schoolers?
This is more or less a ouija board for really lazy people, right?
Just wait until Ashley Madison trains one of these to talk dirty and ask for Venmo transfers.
I watched the heck out of that show back in the day
this is cool. i've often wondered if keystrokes could be read out of an old typewriter accoustically.
I bet you could. The typebars are of different lengths, so they probably have a somewhat different frequency profile. There would be some ambiguity with the bar symmetrically opposite in the type basket, but likely figuring it out would be less difficult than doing the daily cryptogram.
Do the word thang! https://www.youtube.com/watch?v=BMsOKTJNdN8
Fun fact: Some of my private projects still define the symbol LJBAD_DEBUG, which was inspired when I saw the Max Mouse arc of that show when I was a kid.
Obligatory XKCD: https://xkcd.com/130/