This is a completely silly post. The em dashes are mostly used for spacing and formatting, from a brief glance. This does not prove anything about AI usage whatsoever.
Also, a large omnibus bill is not at all comparable to an average bill.
You’re incorrect since it’s a per page metric. Click the links on emdashes - you’ll find I note all nuance. Sorry you aren’t into data, or reading (it’s not for everyone).
Great point - I’d originally wanted to ask it questions after validating for any AI creation. Your point is great validation (and even tho quite different so wtf is up with that) still shows a major overindexing of emdashes, so over 4 emdashes per page vs a result near zero still implies AI use. Thx!
It took me 3 seconds to do. What made you think this was worth posting? Your entire point falls apart because you sent more time justifying using AI which we know just invents stuff, than actually checking it
Which makes your figure for near 0 suspect too. Was that another AI hallucination? Prove it isn't. I don't think you can. Look at the IRA. It has emdashes
When we put things into the world, let's attempt to make sure they are true.
Logan sorry you completely missed the point - you created a low-bound of 3,800 which is 4.x per page and an absurd difference from other bills (at a low bound). You’re correct on my baseline - I’d spot checked bills I could find online from 2025 and 2024 but did not look at the IRA and could not find a use as frequent outside of the table of contents from older bills.
Look at the per-page emdash metric to normalize obv
I don’t know if you searched for double emdashes (which I found in the doc), but since I trust the low bound (ty) and added that with a credit to your comment, while not changing my initial caveats with using an LLM to find the 10.x (I actually thought it would miss some and be too small, with reference to the haystack)
Thx to your low bound, differences don’t matter as much since 4.x as the low bound is an absurd difference (and with all caveats, the point of the post is running 900 pages of AI generated content by Congress with a deadline to presumably dictate reading it w AI, hence my note)
This is a completely silly post. The em dashes are mostly used for spacing and formatting, from a brief glance. This does not prove anything about AI usage whatsoever.
Also, a large omnibus bill is not at all comparable to an average bill.
You’re incorrect since it’s a per page metric. Click the links on emdashes - you’ll find I note all nuance. Sorry you aren’t into data, or reading (it’s not for everyone).
Why would you use an AI to count the number of — rather than just control-f. Control-f brings up 3845 results.
Edited to include reference to first comment ^
Great point - I’d originally wanted to ask it questions after validating for any AI creation. Your point is great validation (and even tho quite different so wtf is up with that) still shows a major overindexing of emdashes, so over 4 emdashes per page vs a result near zero still implies AI use. Thx!
It took me 3 seconds to do. What made you think this was worth posting? Your entire point falls apart because you sent more time justifying using AI which we know just invents stuff, than actually checking it
Which makes your figure for near 0 suspect too. Was that another AI hallucination? Prove it isn't. I don't think you can. Look at the IRA. It has emdashes
When we put things into the world, let's attempt to make sure they are true.
Logan sorry you completely missed the point - you created a low-bound of 3,800 which is 4.x per page and an absurd difference from other bills (at a low bound). You’re correct on my baseline - I’d spot checked bills I could find online from 2025 and 2024 but did not look at the IRA and could not find a use as frequent outside of the table of contents from older bills.
Look at the per-page emdash metric to normalize obv
I don’t know if you searched for double emdashes (which I found in the doc), but since I trust the low bound (ty) and added that with a credit to your comment, while not changing my initial caveats with using an LLM to find the 10.x (I actually thought it would miss some and be too small, with reference to the haystack)
Thx to your low bound, differences don’t matter as much since 4.x as the low bound is an absurd difference (and with all caveats, the point of the post is running 900 pages of AI generated content by Congress with a deadline to presumably dictate reading it w AI, hence my note)