Malignant Intelligence: Prompt engineering and software archeology
summary
Slides are mostly generated with stable diffusion and image models like MidJourney, and the talk shows how a leaked Llama 4‑bit quantization lets large language models run on laptops and even Raspberry Pi. Demonstrations include a dark‑web worm GPT product ($67/yr) and a drug‑design experiment that spits out 40,000 molecules including VX in under six hours. Concrete failures: hallucinations (Mars population, fake citations), Samsung phone moon “enhancement”, adversarial stop‑sign attacks, voice cloning via 11Labs, and reported model drift 98%→2% on simple math.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:24AI as magic controversy
- 7:57AI history and terminology
- 13:24Stable Diffusion local AI
- 21:45AI hallucinations
- 28:11AI ethics and copyright
- 35:22AI vulnerabilities
- 41:00AI in search and tests
- 46:40Regulation and watermarking
- 51:18AI for developers
- 57:14AI sentience and future
related talks
48mMastering the AI Shift: Adaptive Strategies for Software Evolution
43mThe AI Survival Guide: Practical Advice for Stressed-out Security Teams
44mAI Foundations: Mitigate Risks and Boost SOC Efficiency
49mUnderstanding Prompt Injection-Techniques,Challenges and Advanced Escalation
46mMagic of Automation and Everyday Chores
46mScalably Understanding AI with AI
39mLLMs everywhere: The future is now and the past keeps repeating itself
27mAI Demystified Unveiling the Intelligence Behind the Machine