Are LLMs the answer to all our problems?
summary
Claims that LLMs solve everything meet concrete limits: visual outputs show gender and racial bias in training data (DALL·E example), GDPR means personal data used for training cannot be removed, and the EU AI Act bans real‑time public face recognition. Empirical comparisons find many wrong citations and paywalled sources scraped (Perplexity, ChatGPT search, Gemini underperform; Copilot disappoints). Energy numbers quoted: ResNet train ≈ 0.001 t CO2, GPT‑3 train ≈ 570 t CO2 and 1,000 image requests = 200 g CO2; recommends on‑device NER (open source ENS) or smaller models when feasible.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 3:01LLM bias and training data
- 6:10EU AI Act and high-risk systems
- 7:56GDPR vs LLM privacy
- 10:03Copyright in LLM training
- 13:54When generative AI is needed
- 15:38AI model carbon footprint
- 19:55LLM applications and responsible use
- 21:24Management vs engineering reality
- 22:45Innovation vs AI regulation
- 24:34AI law enforcement strategies
- 26:40Future LLM energy trends
- 27:10Q&A
related talks
10mThe Impact of LLMs on the Circular Economy and Beyond
20mLLM: Efficiency and Sustainability in Practice
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
24mHarnessing the Power of LLMs in Advanced Analytics- 12mKeynote: How LLMs Have Changed the Way Organizations Do AI
41mEnhancing LLMs with Graph Technology
17mPsychology of AI
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence