The Dark Side of AI - Environmental & Social Costs Developers Must Know
summary
AI training can need vast resources: GPT-3 training uses about 700,000 liters of fresh water and consumes electricity equivalent to 130 U.S. homes for a year. Model choice and hardware/location matter: Bloom and GPT‑3 differ by a factor of 20 in required carbon-offset trees, and DeepSeek is reported to use ~87% more energy at inference than Yama 3.3. Facial‑recognition bias (Joy Buolamwini example), low‑paid refugee annotators, mineral extraction and e‑waste in the global south, and the advice to reduce, reuse, recycle models make the costs concrete and actionable.
AI-written summary · about
chapters (9)
- 0:00Introduction: Why Talk About the Dark Side of AI
- 0:33Environmental Costs: Energy & Water Consumption
- 1:20Social Impact: Algorithmic Bias & Facial Recognition
- 2:00The Human Labor Behind AI: Data Annotators
- 2:24Global Inequality in AI Benefits and Costs
- 3:01Comparing Model Footprints: GPT-3 vs Bloom vs DeepSeek
- 4:04Conscious AI Usage: Summarization vs Image Generation
- 4:47Sustainable AI: Reduce, Reuse, Recycle Models
- 5:14Final Thoughts: What Developers Can Do
related talks
14mThe Real Cost of a Token: the Hidden Bill of Generative AI
47mThe Rise of AI: A Reality Check on Generative AI
41mThe AI boom: what and why?
01h 00mAI A-Z
54mThe Future of AI: Building the Most Impactful Technology Together
52mBeyond the AI Hype: What's Real, What's Next
27mSustainability in a Buzzing Tech World: Balancing Progress and Responsibility
27mChallenges of Generative AI