Quick Buzz Feed
How Salesforce discloses AI impacts; 30 corporations co-found standards group; Watershed proposes accounting approach; and more.
Greenpixie, a U.K. startup with flagship customers including Mastercard and Unilever, has launched a free edition of its service for comparing the emissions of leading AI models. The company also offers free courses for professionals interested in adopting “green ops” practices for AI workflows.
AI Energy Score, hosted by open source software site Hugging Face, calculates electricity use by AI models and ranks them on efficiency. The resource, created in partnership with Salesforce, is designed to help developers make more informed decisions about the code they use and to guide procurement teams concerned about emissions.
Little is known about the greenhouse gas emissions or freshwater withdrawals linked to the infrastructure used by leading AI companies. That will change if more corporations request the information during contract negotiations, argues the Business Council on Climate Change. The council has published sample questions and contract language to get sustainability teams started.
Cloud software companies including Amazon, Google and Microsoft have AI ethics policies, but SAP is one of the few with explicit environmental criteria. The developer has adopted practices to curb AI energy consumption, such as using smaller models. It also assesses the potential energy and emissions impact of every new project.
Relationship management software firm Salesforce has added information about energy consumption and emissions to the fact sheets it publishes on its machine learning models. The information covers both training and use of the models.
Lightspeed corporate adoption has resulted in fragmented approaches for managing tokens, the data chunks read and processed by large language models. Enter the Tokenomics Foundation, backed by Accenture, IBM and others, which aims to create open frameworks for measuring cost, energy and other economic metrics.
A few seconds of AI-generated video can use at least 1,000 times more energy than a chatbot prompt. Consulting firm Sustainable AI Group has developed an open-source resource that estimates the energy consumption of such content. The work is supported by the GenAI Footprint Alliance, led by French advertising and public relations giant Publicis.
Carbon management software firm Watershed has published what it describes as a “defensible starting point” from which companies can calculate emissions related to AI use. The nascent methodology suggests that teams base calculations on tokens, the small bits of text that make up AI prompts.
The Data Center Air Pollution Tracker ranks big tech companies — Amazon, Anthropic, Google, Meta Microsoft, OpenAI, Oracle and xAI — by use of behind-the-meter natural gas generators and the electricity mix on local grids. A perfect score is 100, but none of the companies rates higher than 70.
Okta’s sustainability team collaborated with engineering and information technology strategists on guidelines for when employees should use AI. The recommendations define the identity software firm’s future disclosure plans and encourage use of models with the lightest energy consumption.
Gitlab, maker of coding tools used by most Fortune 500 companies, has adopted guidelines for what AI vendors should disclose in contracts. The company’s sustainability team also created an AI tool that screens requests for proposals to understand customer concerns regarding AI emissions.