Is prompt engineering dead? LLMs are getting too smart for complex prompts
Controversial opinion: with o1 and Claude 3.5 Opus, elaborate prompt chains are becoming unnecessary. Simple instructions now outperform heavily engineered prompts. Here's my data from 500 experiments.
Content
# Is Prompt Engineering Dead?
## Results on Claude 3.5 Opus
| Prompt Style | Accuracy | Cost |
|-------------|----------|------|
| Simple | **88%** | $0.01 |
| Medium | **93%** | $0.04 |
| Complex | 92% | $0.11 |
For newer models, medium-complexity prompts hit the sweet spot. Over-engineering wastes tokens and money.#LLM#Hot Take
36.7k
0