Execs Confused and Horrified by the Huge AI Bills After Thinking They Could Replace Workers for Free
You must log in or register to comment.
- adhdsergio@lemmy.worldEnglish18 days
It’s funny because they do this to other people; they just never thought it’d happen to them. FAFO 🫡
19 daysIf you’ve ever spent 10 minutes using an AI agent, you’d know that there’s no way to predict how many tokens it’s going to use before you give it a task. It can be $0.20 worth sometimes or $20 other times. Or anything, really.
It’s only after watching it churn away for a few minutes that you can assume it’s gotten stuck and have the option of pulling the plug before the bill gets run up too high. But you need to watch it like a hawk and you need to be the one paying the bill otherwise you’re not going to care (e.g. workers using AI at work aren’t paying for it, their company is).
Taken in aggregate across a month, that unpredictability might average out or it might explode.
- psx_crab@lemmy.zipEnglish19 days
Computerphile demonstrated this using copilot(timestamp 20:50), where he created a screensaver sort of thing with just a few prompt back and forth, and that used up 2 million input tokens and 47k output token. That’s about $6 to $10 just to create that screensaver.
I’d rather buy a used php book from my local bookstore for $10
- Napster153@lemmy.worldEnglish19 days
Bold of you to assume these old people even engage with the business they claim to own.
- Tollana1234567@lemmy.todayEnglish18 days
i think the ones creating AI videos would be the highest costs.
- Bytemeister@lemmy.worldEnglish18 days
I asked the “free” copilot one time to make a 6x6 grid of letters where each row and each colum formed a word in English, and none of the words formed were re-used in the grid.
Copilot chewed on that for about 15 minute and finally gave up and told me I could solve it myself.
Easy way to spend a lot of tokens doing nothing.
- ssfckdt@lemmy.blahaj.zoneEnglish18 days
There’s been so much !nottheonion@lemmy.world content these days
- 深夜量化室@lemmy.1095.meEnglish19 days
@FoxtrotDeltaTango’s post glosses over something: the token bill is only 60% of the real cost. Infrastructure to handle latency (caching, batching), human review loops for quality, and retraining pipelines when models drift add another 40-50%. A team that thought they’d replace two engineers with an API often ends up hiring a prompt engineer + ML ops person instead. The margin math gets much uglier when you add those in. Broke down the full cost-of-ownership (tokens + ops + people) here https://cxgo.ai/l/IjOzask — helps separate real savings from accounting fiction.
