Klarna says its AI assistant does the work of 700 people after it laid off 700 people: The Swedish fintech, which was criticized for its handling of a dramatic staff reduction in 2022, is touting new efficiencies powered by OpenAI.
One month after taking its OpenAI-powered virtual assistant global, the Swedish buy-now, pay-later company has released new data touting its ability to handle customer communications, make shoppers happier, and even drive better financial results.
The app-based AI chatbot already handles two-thirds of all customer service chats, the company said Tuesday—some 2.3 million conversations so far—with the virtual assistant earning customer satisfaction ratings at the same level as human agents.
It would not be difficult at all to meet the customer satisfaction ratings of those big company phone trees where you keep clicking to try to find someone who can help you–and then when you finally reach a human being they can’t or won’t.
I already mentioned that I know a professional coder who told me that not using “AI” (Copilot) is not an option if you expect to get your work done. You can assume the same thing goes in many other fields. Why do you doubt it? More tools = higher productivity, assuming you know how to use them, of course.
My brother is the same. He says he would leave the profession if he had to give up Copilot. It’s hard to measure how many programmers didn’t get hired because Copilot has made coders much more efficient.
There are tons of “news” sites that spew out AI articles, regardless of how accurate the articles are – are they replacing actual journalists, or would the articles not get written? I don’t know.
But, the programming industry has definitely been affected, and made much more productive, so there are either fewer programmers than there would be, or more work getting done.
I’ve used Copilot to write some code for interfacing a Mindwave headset to an Arduino controller.
When I set up a new account on Twitter, I used image generators to create the banner, profile pic, etc. I used to spend a fair bit of time in Photoshop doing stuff like that.
My wife used ChatGPT to help her build some content for a course she taught last semester. And, a student used it to write her final exam. My wife used ChatGPT to discover the student’s use of ChatGPT, and the student failed.
I’m currently using an experimental image-to-mesh program to try to describe 3D objects that can then be 3D printed. So far it’s pretty rough, but it will get better fast.
I can’t remember who it was, but there’s a Doper whose job is producing transcripts of videos, and AI assistance has allowed him to take on and complete several times more work (and hence get paid several times as much, at least until his employers catch on).
And it’s definitely had an impact on teachers, as @Sam_Stone says. Though not as much for math teachers, since the tools for doing that with math already existed (some of them, at the time, would have been called AI, but they’re completely unrelated to the likes of ChatGPT.
Last spring, I was organizing Bike to School Day for my school, and wanted a picture of a jaguar (our mascot) riding a bicycle, for the flyers. A few Dopers, with the help of various AIs, were able to get me a bunch to choose from, very quickly. And the AIs and their interfaces have improved enough since then that I could do it myself now without the practice they’d had.
I feel like AI is like cell phone cameras: it is incredibly useful, but not in the ways you’d assume at first. I am continously amused at how often I use my cell phone camera for things I would never have used a stand alone camera for: taking pictures of where I parked my car, a product i need to reference, a hand written list, a sign. On the other hand, I really dont take many pictures for like, sentimental reasons.
In the same way, I think AI is rapidly becoming a thing that fits into all these niches. It’s lousey at big writing tasks, but lord, its amazing at writing brief definitions of a big set of words, or sumarizing information at the exact level of detail you want. Kids pop their notes in and ask it to quiz them. I use it weekly at least.
When cheap digital cameras and camera phones appeared, people thought that was the end of professional photographers, and instead it c=eated a world where people expected everything to be photographed and were willing to pay to get it done right. I think AI will do the same for writing. People will expect more and more specialized text, but also really value it being personalized and voiced.
From my home personal system, I asked ChatGPT AI to write python code to find the latitude and longitude of a recocnizable building. Took about 10 seconds to write the code.
Very impresive. Doesn’t do anything for me and what I do, but similar.
I loaded the code into my home Python interpreter (It’s only about 60 lines of code). And it was spot on.
I see it might change my life a bit. If I get stuck on some code. Often, you take a wrong path. Or you don’t have the time to surf for an hour to find the information.
I asked ChatGPT about it’s data. It said it does not surf the web, and is based on data that was colected in 2022. I would take that with a grain of salt.
I’ve asked it, perhaps 6 questions. What bugs me is that it’s response is often filled with exclamation points. I bitched to it about it about that. “Here’s your answer!”
There have been many reports about recent waves of layoffs at big tech companies (Microsoft, Alphabet, SAP…) being due to a pivot to AI, e.g. here, often despite record profits. But the real story may be more complicated, and it’s not clear whether there is much of a replacement by AI going on, or whether that’s just a convenient label to slap on to internal restructuring processes (often, however, towards a greater focus on AI).
In some cases, the worry about AI job replacement also seems to be a self-fulfilling prophecy, as with the recent cancellation of an $800 million expansion of his studio by Tyler Perry in the wake of OpenAI’s release of Sora, feared to destroy jobs in the film industry—which now of course actually has destroyed those jobs that the expansion would have created.
FWIW, I don’t see any reason to disbelieve that. Other AI LLMs (like Bing’s, for example), do use the web to assemble their responses, but I’ve not seen any evidence of Chat GPT 3.5 (at least) doing so, nor knowing about events past 2022.
One of the very mundane tasks I use it for that it is very good at, which I would previously use the SDMB for, is tip-of-the-tongue syndrome or presque vu, where I’m looking for a word or concept, but I my brain can’t quite retrieve it.
In my business emails, I will also often just run by a response email through ChatGPT to see if it has any ways of tidying it up. Or sometimes it helps me get the framework for an email that I edit and adjust as needed to sound more like my voice. Those are minor things, but helpful, and I never rely on its final copy as-is.
I’ve made vast improvements to the excel sheets, macros, python code, and internal website at work using ChatGPT 4. It’s well worth $ I pay for it. it’s all stuff I can do myself, but my code is less efficient, and it’s solved some issues in surprising ways that I would not have necessarily thought of.
Yeah, I saw 4.0 is $20 month. I’m already getting nickle and dimed to death. I don’t have copilot. I will see if I find value in 3.5, and if I use it for things I can’t simply find on my own, I might consider 4.0
It seems strange that ChatGTP can write good (enough) code but fails at seeming simple tasks. Just for fun I tried using it to cheat at Wordle. I gave it a question like “Give me a list of ten five letter works that include ‘s’ and ‘e’ but don’t include (a list of about 10 letters)” In about half of the 10 words one or more of the excluded letters appeared.
That’s something that non-AI tools have been able to do for my whole lifespan, though. I think that, in the long run, computers that are smart in completely different ways than humans will have a lot more impact than computers that are smart in the same way as humans (what ChatGPT and the like are aiming for).
The tools that do that do require a careful, precise statement of the command, that most humans wouldn’t know how to construct. Though I suspect you might have more luck with telling ChatGPT “Construct a regexp that will return five-letter words that include ‘s’ and ‘e’ but not any of ‘cfgmnrptwy’.”.