- cross-posted to:
- technology@lemmy.world
- cross-posted to:
- technology@lemmy.world
Sources and leaks from Amazon, Adobe, Atlassian, Citi, and more show what is really happening with AI right now: companies are trying to rein in AI use as costs spiral out of control.
You must log in or register to comment.
Curious to see how it works out when the AI companies need to grow exponentially to be profitable while their biggest customers are cutting back.
I wish, it could turn in to a black hole though 😕
RIP tokenmaxxing
Oh dear! Who possibly could have seen this coming?
Every time I see Anthropic’s logo, all I see is brown starfish.
Is that intentional?
What’s the best way to talk to a customer service chatbot so that it burns a ton of money?
I’m so glad you asked.
Kitboga, a YouTuber who loves to mess with scam callers, has recently made a video showing what you can do to mess with an AI agent.
In a nutshell, they are designed to do what you ask them to do and be WAY more accommodating than any human would be. The more crazy things you prompt it to do that pulls the conversation towards something that no longer resembles a normal conversation, the more they will tend to ignore their original prompts and be easily talked into putting themselves in infinite loops. They also start to spectacularly glitch out because there is no training data for what you’re making them do.
Tons of fun to be had.
AlbuquerqueNewMexicoAlbuquerqueNewMexicoAlbequerqueNewMexico
That was awesome. Anyone could do that.
Most companies will outsource the chatbot to a third party (like Zendesk, as part of Zendesk’s overall support packaging), and those are usually billed per conversation, not per token (at least, not until they update contract terms lol), so you’ll want to make sure you’re starting many many chats and engaging in them (send at least a couple of messages).
Also, some systems will not count against the quota if the AI didn’t “fix” the issue, so always end with “thanks, that helped” to make sure the conversation counts against the limit.
However, if you’re using a product with some copilot crap, ask it for help, and then just constantly ask for minor adjustments or corrections (“no, that’s not quite right, try again. I wanted the color to be a little more blue” “no, try again, now it’s too blue, reduce it”, you get the idea). You’ll burn through soooo many tokens.
If the company using the AI service doesn’t foot the bill because of how their contract with the AI company was set, it is the AI company that loses money. Either is a win to me.
I love you for this comment. That’s all stuff I can do.
surprised pikachu face
Firms that find LLMs useful for this or that would start investing in their own LLM infra likely using Chinese open models. There are now models that perform well in terms of quality and speed that run on workstation-grade hardware. Those are common at corpos that build non-web software.
What I don’t get is why most companies are not already doing this.
I think so many corps are just going on the typical MBA mantra of not owning any infrastructure of their own, because “core competencies” and shiny object FOMO (shiny objects and “we have to scale to Google scale”, LOL) or whatever.
It explains why so many people run shit in AWS even if they don’t really need to and buying hardware and putting it in a co-lo would have made more sense.
This kind of thinking is probably what explains idiots doing huuuuuge AI spends that don’t have to.
I think all of it comes down to short-termism. All the “core competencies” and all that, all it drives to is optimizing for higher profit for the next quarter. And short-termism is built into the system, driven by the necessary competition for capital. I.e. if you need capital for some investment that takes 2 years to materialize but another firm offers better return 1 year from now, capital goes into the second. Run this in a loop over time and you get most firms optimizing for next quarter profts.
E: Why necessary - cause capital is prerequisite to any production, software, service, or hardware.
E2: I’ve thought MBAs are the bane of companies for most of my corporat life. These days I think that’s wrong. They follow the logic of the system and if one does not, they’re being replaced with one who does.
E3: I mean those phenomena you mention are absolutely behind the effects you say. All I’m saying the phenomena are all driven by the basic incentive structure of the system and are largely unescapable in traditional, non-unionized capitalist corpos.
