Humans vs. Machines
summary
Generative systems like GPT‑4, MidJourney and DALL·E produce repeated failures: frequently confident, never in doubt factual hallucinations (wrong arithmetic, invented biographical details), stereotyped biased imagery, and a visual plagiarism problem where outputs are near frame-by-frame copies of copyrighted art and characters. Guardrails are brittle and can be bypassed. Evidence includes side‑by‑side examples of replicated movie and game imagery, a New York Times copyright suit, and experiments showing bigger models increase inadvertent close copying, pointing to licensing and consent as the practical fixes.
AI-written summary · about
chapters (10)
- 0:00The future of AI
- 1:14Current AI limitations
- 4:44Plagiarism and ethics
- 5:53Lack of true understanding
- 6:41Risks and disinformation
- 7:52Power and IP conflicts
- 9:36Visual plagiarism case study
- 13:15Ineffectiveness of guardrails
- 15:39Governance and the future
- 18:52Closing remarks
related talks
37mThe Shadows That Follow AI Generative Models
41mThe Unreliable Computer Revolution - Embracing the Double-Edged Sword of GenAI
57mThe AI puppet dance
59mMirror, mirror: LLMs and the illusion of humanity
30mBattling the Machine: Cutting-Edge Attacks and Defenses in Generative AI
56mSeparating fact from fiction in a world of AI fairytales
27mChallenges of Generative AI
06mInterview with Spela Poklukar