Sort of same thing here. No way I’m even gonna ask innocent questions on a work machine. I will use it on my home machine for some basic quick questions.
Now we hear that that South African dude is going to train his “Grok” model using his own social network. This will make many heads hurt, as X inherited a lot of bots from Twitter. The training of “Grok” will be deeply recursive as it learns from automated message generators and probably will not be able to discern their content from wetware-generated content.
As AI trains itself on AI generated content, and that content continues to snowball, the results will get worse and worse. ChatGPT will be stealing from Grok and Grok will be stealing from Gemini, etc.
Nope. Not once, not ever. Never will, either.
LLMs have “guardrails” to prevent them from doing stupid stuff and to prevent them from being tricked by users. Apparently, these safety mechanisms are extremely easy to defeat. Just put spaces between the letters.
Had a weird design request: The client wanted pictures of middle-aged corporate execs using laptops, all together in one sitting. Since one committee had 6 members and the other 19, he wanted the images to reflect those numbers.
I went through three rounds of image selection. Adobe and Getty tend to stock younger corporate exec images, and pics of them using laptops in groups usually entailed the models standing around one person using the laptop. They all got rejected. Finally, I resorted to Firefly.
In general, we’re supposed to use AI-generated images minimally because they tend to look staged and not wholly natural. I had used touches of AI-generated images here and there to alter backgrounds to eliminate features that stood out too much, and that was as far as I had gone. I fed “A dozen middle-aged office executives using laptops” into the text-to-image field and downloaded about 20 results. I thought I’d get torched, but the client loves them.
I recently sat in for one of Adobe’s webinars about best uses for AI image generation, and it looks like not only is it here to stay, it’s going to get even bigger. GoDaddy and Canva can generate entire websites from text input already, and they include AI-driven data analytics tools to monitor usage. Adobe’s about to release GenStudio, which will enable marketers to generate email and social media campaigns all in one sitting. All they have to do is input headlines, copy text, and the types of images they want, and GenStudio takes care of it. Furthermore, marketers will be able to change copy and redistribute ads in a heartbeat. They have access to AI-generated buying habits of their customer base, and can make ads attractive to individual customers and distribute solely to their emails and browsers.
If you thought privacy issues were bad before, you ain’t seen nothin’ yet.
I’m retired now so it doesn’t really affect me, but after 40 years in communications I believe that AI has the potential to turn the entire creative process into a fill-in-the-blank exercise, and I feel sorry for all the people I’ve mentored over the years on the writing process, all the designers I worked with who labored to turn corporate names into distinctive, memorable logos, all the photographers who labored for hours in bad weather or cramped offices to get the perfect shot,. and so on.
My son recently got married, and he and my daughter-in-law used AI to write their vows. I’m glad I didn’t learn this until after the ceremony, because I would have been very upset. I’d prefer plain-spoken sincerity to machine-generated eloquence.
That probably doesn’t quite line-up with the OP which was ‘have you used it deliberately’. Sure you can split hairs here - “using Google Translate is using AI deliberately” - but I’m not sure how helpful that is to the general conversation, which really does seem to be more about directly engaging with generative AI.
Personally I haven’t deliberately used much generative AI. Not due to any smoldering hostility, but it just doesn’t come up much.
1.) I’m middle-aged, nearing retirement and it has no really useful bearing on my job operationally. It may a bit for design, but I’m involved in that shallowly and mostly at the level of suggestions at the very front end and then end-user testing and tweaking.
2.) As a toy it just doesn’t interest me much. I’m not really into stuff like experimenting with novel images or text. Also doing my own research the hard way is a fun hobby - I like reading books and long articles. I’m generally not looking to increase my efficiency and I have no mass production needs or pressures.
It is highly likely I will slip off the shackles of life without ever having engaged directly with generative AI except as a mostly unknowing consumer of products and services created with it. Or at most shallowly, like using a translation program for a random phrase or article.
Just got this AI summary from a Google search:
A whale’s breach is when it propels itself out of the water, while a breech is when a baby whale is born feet first
I didn’t mention it at the time, but the “Hubble expansion” web page I made in this post was using AI:
It worked almost perfectly right off the bat using this prompt with claude.ai:
I want to make a visualization of Hubble expansion in HTML. The background should be a collection of random red dots fixed in place. In the foreground, but transparent, so that you can still see the red dots, should be a collection of blue dots. The blue dots should be in the same place as the red dots, except moved to be 10% farther from the center. It should be possible to slide the foreground around using the mouse.
It had one bug, which was arguably a bug in my prompt, and which I fixed trivially. However, I did ask it to fix it on its own:
That’s great! It’s almost perfect. The one bug is that when you drag it around, it snaps back to the original 0, 0 centering. I would like it to stay in the new dragged position. I was able to fix the bug myself, but I would like to see your fix for it.
Worked perfectly after that. I did fiddle with the star size/count a bit, but that’s the kind of thing you only know after looking at it. Its default was actually fine on its own.
It would have taken me ~2 hrs to do this since I’m a little rusty with JavaScript. This took under 5 minutes.
AI is amazing for quick things that you want to whip up but that just aren’t quite the mental effort to do yourself.
I remember something I used AI for - last year when we were supposed to come up with our own goals, I used it to come up with some goals, and translate some thoughts I had to business speak
AI is great at business speak.
I used AI a bunch prepping for a scifi RPG campaign I am running in a homebrew setting.
The players are the crew of the first interstellar ship. I was able to generate the rest of the crew pretty easily by feeding the AI a little background info on the setting, then telling it what kinds of characters I wanted.
I told it things like “the ship’s captain is a veteran, she comes from Finland, was in the surface navy for the North Sea Federation, eventually joined the Congress of Nations space fleet and spent a couple decades hunting pirates in the outer solar system. She is very pessimistic and cynical but cares deeply for her crew.” Or “the expedition leader is a biologist who used to be a professor and now heads up a university in Aotearoa, in the Pacific Free States. He is an extremely talented researcher and very trusting, but can be a little naive” or “the chief engineer came to the CN Navy from the Nigeria, in the People’s Union of the Tropics. He is reserved and thoughtful, tending to keep his thoughts to himself.” Or “A journalist traveling with the expedition to report on their progress, she is cybernetically enhanced and comes from the Greater Commonwealth of Sibir.”
For each of the 11 or so NPCs the ship needed, I simply provided a blurb like that and got back a paragraph for each character’s personality, background, strengths, weaknesses, and quirks (a format ChatGPT decided to use on its own, mind you). And literally all of it was great, I gave enough info in the blurbs that nothing that generated went against my vision - it just fleshed it out more.
Just used it this morning. I was looking for a word. I knew it, but it would not float to the top.
Now this was at 4am, I needed to change some code so it would work on another server. This is production server to new production server (basically). So ‘Deploy’ doesn’t quite fit.
Migrate does though.
I recently tested it out by describing a book cover I wanted to use.
It was the ugliest thing I ever saw. Far too busy for a book cover and the airbrushed look made it ridiculous. (Oddly, it metaphorically showed a plot point I hadn’t mentioned to the AI.)
My wife just had me use it to translate a legal contract from Japanese to Chinese.
While it still needed to be checked and verified, it’s so much faster than doing it yourself.
The actual contact is in Japanese so the Chinese is only for the client’s understanding, but it shows the areas which are covered. The client understands that the translation isn’t formal and doesn’t have any legal bearing but it least it gives an idea of what is in the contract. If the client has further questions they can have a human translator check it.
I’m not running a game at present but AI art has become the go-to for everyone when it comes to character portraits. Gone are the days of hunting for a “Ehh, good enough” picture on Google Images. Sure, it took some work to get a good image of a male halfling cleric in leather armor and holding a starknife but I had no chance of finding something that specific on GIS anyway.
Yeah, I plan to make portraits for all these NPCs.
I also generated a decent picture of a six legged alien lizard but I wasn’t as happy with it as I was with the text stuff.
There was a brief heyday of people commissioning art for games. I saw a lot of that online, and there was one time when a player for a game I was running had that done (the one time we had a Zoomer briefly join our group).
On the one hand, I certainly feel for any artist who was making decent money painting portraits for RPG characters on commission. On the other hand, that’s the sort of thing that’s not really accessible to the vast majority of people. I certainly have never felt like I was in the financial position to drop a sizable amount of money on a drawing of my character. I view the fact that everyone now has access to custom art of their RPG character as a bigger benefit than the harm done by the destruction of the incredibly bougie character art market.
And, I’m not even sure the market is destroyed. The sorts of people who value art enough to pay for a custom commissioned portrait are also likely to be the people who want to support artists over AI art and who value the social status of commissioned art.
I had a commissioned piece done for a favorite character and it was lovely but it was also after I had largely finished a campaign with the character and gotten attached. I wouldn’t be willing to pay someone $30-$50+ on a portrait of a character* I might wind up disliking or who might get eaten by a grell in first three sessions. I would still consider a custom piece done for a favorite character I’ve played for a long time but that was always my bar to begin with.
I also thought, when the AI art thing was getting hot, that the ones first hit would likely be the illustrative type artists: character portraits, paperback novel covers, etc.
*I’m talking headshots here; I was never in the market for some $300 full scene image
I pay $20 a month for the thing and just the other night I asked it to create a picture of my wife, in Cossack garb, riding a horse, Moscow in flames behind her (Inna is Ukrainian), and I must say that I unironically just love this picture:
The insane horse. The bear cub, glowering, in a jacket. I have no idea if the clothing is traditional Cossack or traditional bullshit, but whatever. It’s awesome.
Oh, and she doesn’t look like Inna at all, even though I did specify brown eyes. Go figure.