Urgent.News

What's breaking now, across thousands of outlets.

Tech

To COP31: Your Prompts Are Packing Away an Electric Bill

Let's talk about AI Sustainability ahead of COP31. Smarter prompts. Smarter footprint.

To COP31: Your Prompts Are Packing Away an Electric Bill

OpenAI's Sam Altman claimed that the energy spent on ChatGPT prompts amounts to tens of millions of dollars annually. However, the actual energy consumption may be significantly lower. A Google study found that the median energy use for a prompt is 0.24 Wh, although this includes energy used by machines while idle and for cooling purposes.

OpenAI's own estimate is around 0.34 Wh per query, while Microsoft researchers measuring Frontier models got a median of 0.31 Wh. This energy consumption is comparable to running a small town. The environmental impact of AI prompts can be minimized by using shorter prompts and answers, employing smaller models suited for specific tasks, and avoiding unnecessary retries or long responses.

A study of office workers using AI assistants found that halving the number of prompts and answers could reduce energy consumption by 38 times.

Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at hackernoon.com →

More in Tech

Your Gumroad license check probably lets refunded buyers in (and how to fix it)

If you sell software on Gumroad and gate it with license keys, there's a good chance your check looks like this: const res = await fetch ( " https://api.gumroad.com/v2/licenses/verify " , { method : "…

  • Gumroad license check may allow refunded buyers to use software
  • Verify endpoint returns true status even after refunds or cancellations

17,651 subdomains, one server: what checking "is this job still open?" taught us

I run Job Opportunities API (JOA), a jobs API that only reads employers' own career sites and applicant-tracking systems. The hard part is not finding jobs. It is knowing when a job is gone.

  • JOA's checker treated each subdomain as separate host, causing 17,651 requests per second
  • JOA now groups hosts into lanes based on server that answers, not hostname
  • JOA distinguishes between 410 (gone) and 429 (rate limit exceeded) status responses

More from Tuesday 20 October →