AI Showdown: The Good, The Glitchy, and The Risky
summary
Generative vs discriminative models are practical mappings that lose information: ChatGPT creates text prompts that DALL·E and Adobe Generative Fill turn into images but often produce unexpected artifacts, requiring chains of tools to get a usable result. The talk shows concrete risks: AI-generated polymorphic malware and convincing Frank Sinatra deepfakes, a simple poisoned-model trigger that flips behaviour in 2024, and malicious PyPI packages that run code on install. Recommendations point to vetting provenance, using Hugging Face or local Llama experiments, and orchestration with LangChain + LlamaIndex alongside SSF scorecards.
AI-written summary · about
chapters (15)
- 0:00Welcome to AI Showdown
- 1:27Common AI misconceptions
- 5:24Generative AI text mapping
- 9:41Challenges in AI image generation
- 16:42AI in healthcare and finance
- 18:05AI cyber threats and deepfakes
- 22:38Poisoned AI models and risks
- 27:17AI’s impact on software supply
- 33:00AI in software development
- 36:29AI infrastructure and Kubernetes
- 40:29Getting started with generative AI
- 43:21AI code assistants
- 47:58Security risks in AI libraries
- 51:05LLM frameworks for Java
- 52:29Human oversight in AI training
related talks
37mThe Shadows That Follow AI Generative Models
27mChallenges of Generative AI
13mNavigating Generative AI and Its Risks
32mSecurity Implications of Artificial Intelligence Synthesizers
45mOverview of Risks, Vulnerabilities and Paths to Safer Use of Generative AI
52mAI Is Just Software, What Could Possibly Go Wrong?
41mThe Unreliable Computer Revolution - Embracing the Double-Edged Sword of GenAI
28mEthical AI?!