📫 Subscribe | 🗣️ Hire us to speak | 🤝 Sponsor this newsletter | 🎓 Events

Welcome to the Friday edition of our newsletter. We spend Fridays going deeper into tools and trends related to generative AI (and Tuesdays sharing news updates). This week, Professor Snider discusses how to lessen your energy footprint while using AI tools.

You've seen the headlines. Every time you ask an AI tool a question, you're supposedly putting another strain on the power grid.

For some people (including several of my undergrad students), that’s enough to oppose using AI altogether.

I understand where they are coming from. But that’s a little like refusing to turn on a single light in your house because it uses electricity. Sitting in the dark isn't the smart move. Flipping the switch off when you leave the room is.

You can use these tools (and it’s hard to envision my students getting a job where they won’t) and still be thoughtful about the energy behind them. Here are five habits that actually help.

TLDR is the free daily email with summaries of the most interesting stories in startups, tech, and programming. The stuff worth knowing, minus the doomscrolling.

Issues are curated by ex-Google and Anthropic engineers and land in your inbox before your morning coffee. A 5-minute read, and you walk into the day already knowing what your team is still catching up on.

Tech is just the start. We also cover AI, marketing, dev, and more. Pick the briefs that match your work.

Free, daily, and read by 7M+ subscribers. Subscribe and let the experts do the digging for the tech news that matters.

Subscribe for Free

The big scary stat that one AI prompt uses ten times the energy of a Google search is mostly outdated. It came from a 2023 estimate built on old hardware.

The newer numbers are smaller. OpenAI's Sam Altman has said the average ChatGPT prompt uses about 0.34 watt-hours. That's roughly what an oven pulls in a second, or an LED bulb running for a few minutes. A 2026 UNESCO report landed in the same neighborhood.

One prompt is tiny, but the real cost is scale. More than a billion people now use these tools, and a lot of those prompts are long, heavy, and wasteful.

That's good news, actually. If the cost lives in scale and waste, then small habits add up. Here's where to start.

Not every question needs the biggest, smartest model in the room. Fixing a typo, summarizing a paragraph, or rewriting an email doesn't require a heavy reasoning engine grinding through its full process.

The reasoning models (the ones that "think" before they answer) use far more energy than the standard ones. Save them for the hard stuff. Use the lighter, faster option for the everyday stuff.

This isn't a fringe idea. That same UNESCO report found that matching a smaller, task-specific model to the job can cut energy use by up to 90% without losing quality. Most tools let you pick your model. Pick down when you can.

Most wasted energy comes from do-overs. You write a vague prompt, get a vague answer, try again, tweak, try again. Five prompts to get one decent result. Every one of those runs costs energy.

A clear prompt up front gets you there in one pass. This is exactly why Chris Porter and I built the BRIEF method – Background, Role, Instructions, Examples, Format. Give the model what it needs the first time and you stop burning prompts chasing the answer.

Better prompts aren't just greener. They're faster, and they make the tool work better for you. That's a rare win-win.

Length is where energy quietly balloons. A short question is cheap. Pasting in a 200-page document is not.

Researchers have measured this. A quick prompt sits under one watt-hour. Attach a long article and you're around 2.5. Feed in a 200-page document and you're closer to 40 watt-hours for a single request.

So be intentional. Paste in the section you actually need, not the whole report. And tell the model how long you want the answer – "give me three bullet points" instead of letting it write five paragraphs you'll never read.

Every time you start fresh, the model has to rebuild context. Batching related questions into one thread is more efficient than firing off a dozen separate chats.

Think it through before you start. Ask your follow-ups in the same place. It's tidier for you and lighter on the back end.

This is where I agree with my non-AI students. Often we don’t need AI. You don't need AI to add numbers, check a date, or look up a fact you could find in two seconds elsewhere. A calculator, a search bar, or your own brain will do it for a fraction of the cost.

AI is incredible at the things it's actually good at. It's overkill for the things a simpler tool already nails. Knowing the difference is most of the battle.

The answer to reducing your AI power usage is to use it well – the same way you'd flip off the lights on your way out the door. Small, automatic, and it adds up across a billion people doing it too.