Human-written demonstration (OpenAssistant volunteers)· OpenAssistant (human-written)
OpenAssistant Conversations (oasst2)
can you write haiku for different inductive biases in deep learning models? start with…
In OpenAssistant data, “assistant” turns were written by human volunteers as demonstrations. They are not model outputs.
- User
can you write haiku for different inductive biases in deep learning models? start with attention
- Assistant
Attention bias shines, Focusing on important parts. Deep learning insight gains. Transformer bias gleams, Sequential information flows, Language is its dream.
Convolution bias sparks, Spatial structure it hones, Images it marks.
Recurrent bias weaves, Sequences it processes, Memory it retrieves.
Generative bias dreams, Data it generates seems, Real, yet not it seems.
- User
Now do it as a rap
- Assistant
Sorry, you are right. Here is haiku Machine bias shines bright Complex tasks it can do Insight it brings light
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →
- Human-written demonstration (OpenAssistant volunteers)Compare side by side →