Technology
This is the official technology community of Lemmy.ml for all news related to creation and use of technology, and to facilitate civil, meaningful discussion around it.
Ask in DM before posting product reviews or ads. All such posts otherwise are subject to removal.
Rules:
1: All Lemmy rules apply
2: Do not post low effort posts
3: NEVER post naziped*gore stuff
4: Always post article URLs or their archived version URLs as sources, NOT screenshots. Help the blind users.
5: personal rants of Big Tech CEOs like Elon Musk are unwelcome (does not include posts about their companies affecting wide range of people)
6: no advertisement posts unless verified as legitimate and non-exploitative/non-consumerist
7: crypto related posts, unless essential, are disallowed
view the rest of the comments
I’ve worked with AI companies where their “AI” is mostly just outsourced to some human in a low wage country. Some small part was actual AI algorithms, so they weren’t completely lying to investors.
Setting up actual AI is expensive and time consuming, especially for some new startup. Often it involves creating large amounts of labeled data and even that might not be enough. There are so many uncertainties involved. It might only be possible to achieve 90% or even lower accuracy for a given problem with actual AI algorithms, so the quick and easy solution is to outsource.
I believe this practice is more common than most think.
Sure, but the fact that not all AI isn't really AI doesn't mean it isn't real. I run local LLMs on my home computer to perform various tasks, I can shut off my Internet entirely and they still work. There isn't some secret line out to a third-world sweatshop where outsourced labor is frantically typing responses to the thousands of queries generated by my scripts.
Training an AI is expensive and time consuming, but simply using one can be very straightforward.
I get you. It’s just that when a company says they’re offering AI solutions, it’s most likely just outsourced workforce under the hood.
Companies can either spend one or two years of R&D, possibly involving PhDs, to solve a task using AI, or they can get the product out to the market in a few months by outsourcing the AI part. The first option might not even be up to par with the outsourced option even with all the money spent on R&D.
You're still talking about training AIs, though. Using AIs doesn't require years of work and PhDs to research. You just sign a contract with one of the AI service providers and they give you an API. You may need to do a little scripting to hook up a front end and some fiddling with prompts and parameters to get the AI to respond correctly, but as I said above, I've done this myself in my own home. Entirely on my own, entirely just for fun. It's really not hard, I could point you to a couple of links for some free software you could use to do it yourself. Heck, even the training part isn't hard if you're starting with one of the existing open models and you've got the hardware for it.
Do you really think all those companies out there with chatbot "help staff" (that speak perfect English and respond faster than a well-trained typist could type) are most likely just outsourced workforce to some cheap foreign company? What is the hundreds of billions of dollars worth of computer hardware the AI service providers are running actually being used for, if not that?
I’m talking about the entire process from design to product. Ok, maybe those useless chatbot “help staff” might be actual LLMs, but that Amazon grocery store used as example in the article was just Indian labor all along.
As soon you want to solve a very specific problem using AI, it can quickly get time consuming and expensive to develop the product. Maybe that off the shelf AI model isn’t good enough for your particular problem? Maybe it only gives 75% accuracy when you really need 95% to be competitive in the market. In that case you need to compare different models, figure out if there’s any trick you can do to boost the accuracy, try out different training strategies, etc.
And once the model has 95% accuracy on your own labeled data, it might turn out it’s completely worthless out in the field because it turns out the data you collected isn’t representative of the reality.
At that point you might just try to figure out how to offload the work someone else. I’ve even heard of self driving car companies who did exactly that.
Going back to my original comment:
The fact that Amazon was faking it in this one instance doesn't poof all the actual AI out of existence. There are plenty of off-the-shelf AI models that are good enough for various particular problems, they can go ahead and use them. You said it yourself, the chatbot "help staff" might be actual LLMs.
As I said, most companies using AI will likely be hiring professional AI service providers for it. That's where those hundreds of billions of dollars I mentioned above are going, where all the PhDs spending years on R&D are working.
The current AI bubble started around february of last year when every C-level of every major company entered into a mass hysteria about being left behind if they didn't integrate LLMs into their service. This is different from the zeitgeist that came before it.
I can confirm this.