freshness blog post #29
Conversation
_posts/2025-03-5-freshness.md
Outdated
|
|
||
| Given a prompt at submitted at time $$t$$, we examine the following: | ||
|
|
||
| - Nearest neighbor with all prompts submitted before time $$t$$ |
There was a problem hiding this comment.
We may want to make this clearer, e.g.
- The largest similarity between the current prompt and any prompt submitted before time
$$t$$ - The largest similarity between the current prompt and any prompt submitted at least one day before time
$$t$$ - ...
_posts/2025-03-5-freshness.md
Outdated
|
|
||
| ## How do we measure prompt duplicates? | ||
|
|
||
| Prompt duplicates are measured by the cosine similarity of the text embeddings (OpenAI's text-embedding-3-small). If the similarity between the embeddings of prompt a and prompt b are greater than or equal to 0.7, we consider it a duplicate. This threshold is set by manually looking through examples to determine when two prompts are asking the same thing. A random sample of prompt pairs with their similarities are provided on our hugging face. |
There was a problem hiding this comment.
Hugging Face? HuggingFace? Don't know what's the right way to capitalize/space this lol.
_posts/2025-03-5-freshness.md
Outdated
| } | ||
| </script> | ||
|
|
||
| While we do see a downward trend in proportion of unique prompts over time, this decrease is plateauing. Interestingly, we also see certain dates where prompt freshness is significantly lower than neighboring dates: we will get to why that is in the next section. |
There was a problem hiding this comment.
Can we say something a little more descriptive here? E.g.
"If you look at the above analysis, the proportion of fresh prompts decreases as a function of
However, as
No description provided.