I’ve used Claude before to extract text from high-quality images but this is the first time I’ve tried with low-quality images. And it worked very well.
I just took a screenshot from a video podcast on YouTube, which I wanted to keep in my notes.
Then I asked Claude to recreate it: “Create this table in markdown”.
| Human Expert | AI | Notes | |
|---|---|---|---|
| Breadth | + | +++ | |
| Depth | +++ | ++ | |
| Breakthrough Insight | +++ | + | Humans’ biggest edge |
| Speed | + | +++ | ~10X faster (writes faster than people read) |
| Cost | $$$ | $ | ~10% cost |
| Availability / Parallelizability | + | +++ | AI’s biggest edge |
| Time Horizon / Memory | ++ | + -> ++ | Long-term, Episodic, Procedural, Working Memory – Oh My! |
| Technology Diffusion Speed | + | +++ | See: “Natural Selection Favors AIs over Humans” |
| Bedside Manner | ++ | +++ | When Doctors Use a Chatbot to Improve Their Bedside Manner |
| Adversarial Robustness | +++ | + | Much easier to confuse/trick AI |