Interview with Spela Poklukar
summary
Daily use of GPT yields a roughly ~50% hallucination rate that drops for popular libraries but spikes for niche tools (example given: MLflow). Retrieval and grounding using Retrieval Augmented Generation and fine tuning reduce but do not eliminate false outputs, so expect occasional failures during coding or cloud setup. Bias mitigation relies on curated training data and moderation, and generative models raise a policy question about who captures the productivity gains.
AI-written summary · about
chapters (4)
- 0:00Start
- 0:05What is the most extreme level of hallucination that you have experienced while working with
- 2:57What do you think is important to do to eliminate bias from AI models?
- 4:49Do you see any other challenges in the realm of AI in general?
related talks
35mChatGPT is lying, how can we fix it?
41mThe Unreliable Computer Revolution - Embracing the Double-Edged Sword of GenAI
32mSecurity Implications of Artificial Intelligence Synthesizers
27mChallenges of Generative AI
46mConversations with ChatGPT: illusion or reality?
45mPanel Talk
32mWhy do LLMs hallucinate?
41mTruth Checker: Generative Large Language Models and Hallucinations