ai-energy

Strel connects electric vehicle owners with private charging point hosts. Find a charge anywhere, or monetize your own station. #GreenTech #EV". Ajoutez aussi le lien du site web ici s'il ne l'est pas déjà.

Adds a banner to Claude and ChatGPT with an estimated token count, flags when a chat is getting too heavy, and helps you summarize and start fresh in one click — with an estimate of the processing and energy you'll save.

Estimate the water, energy, and CO2 footprint of your Claude Code usage from local transcripts. Read-only offset-cost estimate plus a self-contained infographic. Portable, offline Claude skill.

Should space power AI by beaming energy down or by hosting GPUs? An open, falsifiable model: quote-per-claim evidence ledger, derivations, paper and an open photonic test chip. Built with AI agents.

Claude Code plugins that make the environmental cost of AI visible. Friendly Reminder shows the water behind every response.

Put AI compute in the 82 million houses that already have a grid connection. An exact-arithmetic feasibility study of residential AI inference as an alternative to the data center buildout.

Interactive scenario study: if AI decisions got radically cheaper, would total AI inference energy fall or rise? A Jevons-paradox model with every number cited or labeled as an assumption, a Monte Carlo, and an independent Python check. A scenario tool, not a forecast.

SOMA AI Emission Factors (SAEF). Open CC BY 4.0 carbon and water factors for AI inference, per million tokens, by model class and cloud region. Companion to arXiv 2606.10660, under peer review.

A PLAN TO WIN THE AI WAR AND POWER A NATION

Estimation dashboard for the carbon cost of campus AI use at Northeastern. Models AI energy draw from published coefficients and stated assumptions.

Estimates the CO2 footprint of OpenRouter-visible LLM inference — estimates with uncertainty ranges, not measurements.

An order-of-magnitude bound on the electricity used by automated web traffic, and an account of why public data supports no single figure. Two methods, both reported, disagreeing by a factor of 5.3 around roughly 10 TWh per year. Python 3, standard library only. DOI 10.5281/zenodo.20512703