Ever used voice-to-text? Or any voice recognition at all (Alexa, Siri, etc.)? Google Translate? Played computer chess? Searched for images on the web? Or your own pictures on your phone? Or just taken a picture at all on your phone?
If you’ve done any of these things, then you’ve used AI. And specifically, modern neural-net based AI.
Perhaps you’ll reply: “Oh, I meant AI like ChatGPT and those image generators. The ones scraping up everything on the web. Plagiarism engines.”
Well, all those other applications have already been doing all that. The language translators were trained on other people’s data. When you do an image search, the AI was trained on vast amounts of public imagery, just like the image generators. Even the chess engines were trained on huge public databases (though in that case at least, we do have engines that were trained by playing against themselves). And so on.
You might be underestimating just how pervasive AI is at this point, unless you’re totally disconnected from modern technology (unlikely given that you’re posting here). Generative AI is a step above what we’ve seen before, but it’s not fundamentally different either in its implementation or in where they source the data.
Create CSV tables from graphics. This is a godsend for me, quite frankly.
Write simple memos and letters. Yeah, I can do it myself, but why bother spending 15 minutes when I can take 3?
Translate English into Ukrainian. One sentence at a time… and Inna had to correct some mistakes… but it was a lot quicker for her to correct bad Ukrainian than to compose it directly (her Ukainian is pretty rusty, mine non-existent.)
To me, it’s like people arguing that arithmetic done on a calculator is not ‘real addition’ because it doesn’t involve putting apples next to one another.
I recently retired, so no. I played with it a little before i retired, but my field is fairly obscure (actuary) and it wasn’t there yet. I don’t find it onerous enough to write a simple email for it to be worth firing up AI, especially as I’d still have to edit it.
That sounds useful. But when i quit, i was an Excel expert, and i could have answered your question in plain English, too. (I’m fact, when i worked on a joint volunteer project, i DID give those answers.) When i wanted to do something outside my expertise, i found Google was my friend.
But i think AI has a ton of potential, and i encourage younger actuaries to learn to use it.
A recent update to Win 11 installed Copilot in my taskbar. This was quite unbidden, so I asked it how to turn itself off. It happily gave me the answer, which I followed. Now my Windows is now longer infested with Copilot. This marks the first and only time I deliberately used AI.
Yes and no.
It’s very useful for small things – it’s batting at 100% as a thesaurus (it’s always successfully guessed the word I was struggling to remember).
But for me, it sucks at coding right now. Any code I’ve asked it to write has either been simply useless or required so many manual fixes that it didn’t really save me any time. This is with the free chatGPT though; I would be interested to try the better, paid version.
The free version was last updated in January 2022. I’ve actually hit a road block or two when it frankly told me it had no idea what I was talking about. Stuff that didn’t exist in 2022.
I’m sort of thinking about the paid for version. It’s eleven cents a day.
I use Adobe Firefly in graphic design occasionally. Adobe Photoshop, Illustrator and InDesign have beta AI generators as well.
Our department has talked about proper use, and we’ve come to understand we shouldn’t use it to mimic an established art style, one that could already be trademarked. We already avoid choosing images that look staged or unnatural, and there’s now plenty of AI-generated images available in Adobe Stock and Getty Images. They’ve already filtered their AI content to avoid copyright/trademark issues, so they get the blame if we use an image that violates such rules.
There have been a few times when I’ve used Firefly for text-to-image generation. Once was when a project wanted anniversary certificates with big blocky numbers that had previously been made with 3D software. I just used an old image as a reference and got Firefly to generate the big blocky numbers on a brightly-lit stage. It worked and the client didn’t ask for any revisions.
Another time, I had been collecting images to use in an article about how some governments use AI for research and tracking. Our brand uses DEI guidelines, and in one case I had an image of a woman using a cell phone. It was a close-up of the phone, so only her fingers were visible, and she was white. The art director asked for a black woman, and Adobe Stock didn’t have a black equivalent of the image that I could find. So, I used Firefly, and not only did it generate a authentic-looking image, it added little personality touches like a wristband and glossy fingernail polish.
I also made a series of email banners for clients in Pakistan. There were three images of college-aged young Central Asian professionals, and one of them was AI-generated. The client asked for one of the images to be replaced, and it wasn’t the AI image.
Prospect lists for the business. We target franchisees, and the single best source of contact information are the lists of franchisees in each franchises Franchise Disclosure Document.
I used to copy and paste data from PDF’s into Excel, and… hopefully… I get a bunch of mistabled data which I would have to then clean up. Took forever
No longer. Now I just do a quick screenshot of the table, put it in CGPT, and ask it to convert it to a CSV. Still not perfect as I can only do a screenshot at a time, but it’s so much faster.
“Massive” is a relative term. AI now makes up a decent chunk of the computing power we use. This makes perfect sense, since it is now a significant chunk of the computing we do.
It’s one thing to say that ‘AI uses as much power as a small country’, as many headlines recently declared; but we should remember that there are 200 countries in the world, some of which are decidedly not small, and it is in those less small countries where the bulk of our species’ collective industrial activity takes place, and those activities are what drives our energy usage.
As noted above, AI usage will become more efficient the more we do it. I don’t think energy usage will decrease much, because most likely the amount of AI computation we do will increase exponentially even as the cost of computation decreases.
But even if AI energy usage increased fivefold in the next decade, it would still be dwarfed by the things we do that actually consume a lot of energy, like the production of massive amounts of physical things and the shipment of those things all over the planet. I don’t think we’re going to slow down or stop that process, either; in fact, I would hope we increase it exponentially, so that more and more people in more and more countries can reap the benefits of our modern society.
Which means we will need enormous amounts of energy in the future, and given the devestating impact of burning fossil fuels, that energy will need to come from renewable sources. A lot of people are going to have to work very hard over the next couple of decades to make that a reality. Luckily, advancing AI will act as a force multiplier for those efforts.
As a VFX artist, I use it for image generation fairly regularly. I use it to generate the base texture sheets for the effects I make. It saves a ridiculous amount of time.
I’ve also used AI to help me write my annual self-reviews, because self-reviews are stupid.
Nope. When ChatGPT first appeared, I played around with it a bit to see if it looked as if it could really pass a Turing test.
I found it was easily fooled. It is very good at generating plausible bullshit… but there’s quite enough of that around already in corporate environments. Who needs more?
I will admit I haven’t tried it for code generation.
Haven’t tried it for conversation/Turing test. No point. But if you need a snippet of code for something, or a direction to proceed, it seems pretty good.