Thought for the day:
There are four stages to for humans to master any skill or capability
- Unconscious incompetence
- Conscious incompetence
- Conscious competence
- Unconscious competence
However, AI has no consciousness, it is either Incompetent, i.e. provides the wrong outcome, or Competent, provides the right one.
Whether right or wrong, it is confidently right or wrong. We see this in humans often too. The Dunning-Kruger effect is when the less competent someone is, the more confidence they exhibit.
Reasoning models like OpenAI o3-Mini or DeepSeek R1, go some way via inference training to mitigate this lack of awareness. But unlike humans an AI’s competence doesn’t increase from session to session, i.e. with practice.
When architecting a GenAI solution, whether a RAG system, or an Agentic platform, consider an way to build in feedback – this could be another AI agent who’s job it is to critique an outcome, even a process outcome, and save this in the platform instruction set to mimic improvement.
Remember “human in the loop”
The other thought on the four stages of competence is a question.
“Do we need to develop mastery to create, or using AI, could we create even without having first mastered a skill?”
E.g., Could I create a song using Suno.com or Udio.com, that would evoke an emotional response, without any musical skill or knowledge? Or do I need to understand the structure of music, and how to create music first, to generate something that has impact and “soul?”
Would I create a better song, even with GenAI tools as an accomplished musician? Or is everything created with AI just good for prototyping and “soulless?”
Tool for the day:
I’m studying cardiology currently and use NotebookLM to create Overviews, Study Guides, and Audio Podcasts for revision.
I am still blown away by how “human” the voices sound, the level of energy and humour they bring, and the analogies and metaphors they create to help remember complex concepts.
If you are researching or studying anything, I strongly recommend NotebookLM – oh yes, entirely free.
On the Horizon?
Manus AI is topping the hype charts this week.
Think of it as the baby of OpenAI Deep Research meets OpenAI Operator with the brain of Claude Sonnet 3.7. An “independent” tool that can operate a browser and complete deep research with reasoning, performing actions on your behalf. From my research to date, it appears to be over hyped, yet still very competent.
Also, I am excited for the nVidia GTC conference next week. More about what I’ll be watching next Monday.
No Kill Switch 
Leave a comment