3 ms·
Does this also mean giving better system prompts that encourage this behaviour also substantially help?
by glass_door 2y ago
Does this also mean giving better system prompts that encourage this behaviour also substantially help?
- astrange 2y agoIn my experience models aren't very good at following such prompts. Smart "non-reasoning" models like Claude 3.5 could, but would generate so much text when thinking they ran out of context.