Reposted by (dialupready)
do you think Vance read that book before or after marrying his Indian wife and fathering Indian children?
disgusting evil book, the fact that Indian-American right-wingers overlook this shows how self-hating and/or cucked they are
disgusting evil book, the fact that Indian-American right-wingers overlook this shows how self-hating and/or cucked they are
November 14, 2025 at 12:29 AM
do you think Vance read that book before or after marrying his Indian wife and fathering Indian children?
disgusting evil book, the fact that Indian-American right-wingers overlook this shows how self-hating and/or cucked they are
disgusting evil book, the fact that Indian-American right-wingers overlook this shows how self-hating and/or cucked they are
the United Kingdom is also doing that
both countries perma-fucked by voting in moderate conservatives (the most effective agents of Satan) with Merkel and Cameron
both countries perma-fucked by voting in moderate conservatives (the most effective agents of Satan) with Merkel and Cameron
November 13, 2025 at 11:18 PM
the United Kingdom is also doing that
both countries perma-fucked by voting in moderate conservatives (the most effective agents of Satan) with Merkel and Cameron
both countries perma-fucked by voting in moderate conservatives (the most effective agents of Satan) with Merkel and Cameron
That’s true, it taps out at PS1, but I don’t think there are file permissions issue
November 13, 2025 at 9:17 AM
That’s true, it taps out at PS1, but I don’t think there are file permissions issue
the Passenger: Not Good
you don’t want to work through that meandering book to get the ending you’ll get, it’s just not worth it
you don’t want to work through that meandering book to get the ending you’ll get, it’s just not worth it
November 13, 2025 at 5:59 AM
the Passenger: Not Good
you don’t want to work through that meandering book to get the ending you’ll get, it’s just not worth it
you don’t want to work through that meandering book to get the ending you’ll get, it’s just not worth it
no good I tried this it gives OCR the same (bad) quality as the one already in TEXT/, I suspect that was already generated with tesseract
its because of how the images are (censor bars, bad angles, missing text, odd cutoff points)
has to be a AI model for this some news org should find the money
its because of how the images are (censor bars, bad angles, missing text, odd cutoff points)
has to be a AI model for this some news org should find the money
November 13, 2025 at 5:12 AM
no good I tried this it gives OCR the same (bad) quality as the one already in TEXT/, I suspect that was already generated with tesseract
its because of how the images are (censor bars, bad angles, missing text, odd cutoff points)
has to be a AI model for this some news org should find the money
its because of how the images are (censor bars, bad angles, missing text, odd cutoff points)
has to be a AI model for this some news org should find the money
30 images took 8 credits through, I can’t do it on grad student money
The number you have to hunt through TEXT/001 has IMAGES/[001-005]
The number you have to hunt through TEXT/001 has IMAGES/[001-005]
November 13, 2025 at 4:33 AM
30 images took 8 credits through, I can’t do it on grad student money
The number you have to hunt through TEXT/001 has IMAGES/[001-005]
The number you have to hunt through TEXT/001 has IMAGES/[001-005]
there’s a somewhat relation between the IMAGES and the TEXT filenames, the number at the end somewhat corresponds to which few images will be OCR’d in that text.
it’s why i was trying the Deepseek OCR thing, trying to have each IMAGE file and folder have a corresponding text file with better OCR
it’s why i was trying the Deepseek OCR thing, trying to have each IMAGE file and folder have a corresponding text file with better OCR
November 13, 2025 at 4:32 AM
there’s a somewhat relation between the IMAGES and the TEXT filenames, the number at the end somewhat corresponds to which few images will be OCR’d in that text.
it’s why i was trying the Deepseek OCR thing, trying to have each IMAGE file and folder have a corresponding text file with better OCR
it’s why i was trying the Deepseek OCR thing, trying to have each IMAGE file and folder have a corresponding text file with better OCR
none of the documents are collected into the separate books/documents that they are, there’s no link between anything just random folders
November 13, 2025 at 4:27 AM
none of the documents are collected into the separate books/documents that they are, there’s no link between anything just random folders
if you have GPUs or Colab credits could you run the whole image set through Deepseek OCR? The GOP committee published OCR is really bad and no filename link with respective images
I got it working but 30 images took 8 credits. Maybe it doesn’t need a A100-level model but it does need better OCR.
I got it working but 30 images took 8 credits. Maybe it doesn’t need a A100-level model but it does need better OCR.
November 13, 2025 at 4:26 AM
if you have GPUs or Colab credits could you run the whole image set through Deepseek OCR? The GOP committee published OCR is really bad and no filename link with respective images
I got it working but 30 images took 8 credits. Maybe it doesn’t need a A100-level model but it does need better OCR.
I got it working but 30 images took 8 credits. Maybe it doesn’t need a A100-level model but it does need better OCR.
this is an optional ui for retroarch, I don’t use it now since it’s not too happy with touchscreens i just use the default skin
November 13, 2025 at 4:17 AM
this is an optional ui for retroarch, I don’t use it now since it’s not too happy with touchscreens i just use the default skin
no Retroarch is an official App Store listed App now Apple relented with emulation (what’s missing is the BIOS files of the consoles needed for the emulator cores). Just playing Xenogears for now
November 13, 2025 at 1:41 AM
no Retroarch is an official App Store listed App now Apple relented with emulation (what’s missing is the BIOS files of the consoles needed for the emulator cores). Just playing Xenogears for now
just getting these files was 8/100 compute units in Colab Pro A100s are expensive
November 13, 2025 at 1:00 AM
just getting these files was 8/100 compute units in Colab Pro A100s are expensive
Nothing was properly documented for this, needed to run inspect on model.infer to stop it from writing bounding boxes (though if people have the storage, they can add <|grounding|> back to the prompt and have save_results=True to save the bounding boxes images and multimarkdown files)
November 13, 2025 at 12:59 AM
Nothing was properly documented for this, needed to run inspect on model.infer to stop it from writing bounding boxes (though if people have the storage, they can add <|grounding|> back to the prompt and have save_results=True to save the bounding boxes images and multimarkdown files)
I think i have a better OCR here, but don’t have the credits to fully go through it all
bsky.app/profile/dial...
bsky.app/profile/dial...
I got a working colab to go through the Oversight drive and OCR to text with Deepseek-OCR. The script text is topaz.github.io/paste/#XQAAA...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
paste
topaz.github.io
November 13, 2025 at 12:54 AM
I think i have a better OCR here, but don’t have the credits to fully go through it all
bsky.app/profile/dial...
bsky.app/profile/dial...
I got a working colab to go through the Oversight drive and OCR to text with Deepseek-OCR. The script text is topaz.github.io/paste/#XQAAA...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
paste
topaz.github.io
November 13, 2025 at 12:51 AM
I got a working colab to go through the Oversight drive and OCR to text with Deepseek-OCR. The script text is topaz.github.io/paste/#XQAAA...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
you can compare the output for the files in IMAGES/004 in this zip (the cmte’s OCR is under TEXT/001 with similar file names) send.vis.ee/download/cac...
the ocr here is pretty bad (from what i saw in the google drive)
maybe only so much you can do
maybe only so much you can do
November 12, 2025 at 11:10 PM
the ocr here is pretty bad (from what i saw in the google drive)
maybe only so much you can do
maybe only so much you can do
I tried for some of the email images I mean, it does arrange the headers and the content in the text output quite well
November 12, 2025 at 11:05 PM
I tried for some of the email images I mean, it does arrange the headers and the content in the text output quite well
I tried a publicly available deepseek-ocr endpoint www.alphaxiv.org/models/deeps... and the output is quite impressive, but im incompetent, can’t figure out how to get this to work on colab(mounting the oversight cmte google drive shortcut or getting deepseek ocr to work in huggingface transformers)
November 12, 2025 at 11:04 PM
I tried a publicly available deepseek-ocr endpoint www.alphaxiv.org/models/deeps... and the output is quite impressive, but im incompetent, can’t figure out how to get this to work on colab(mounting the oversight cmte google drive shortcut or getting deepseek ocr to work in huggingface transformers)
ipad mini
this would be a better experience on Android, the controller (gamesir x5) has a special Android mode where it can control features like the volume and apps get more access to things like the filesystem
this would be a better experience on Android, the controller (gamesir x5) has a special Android mode where it can control features like the volume and apps get more access to things like the filesystem
November 12, 2025 at 10:57 PM
ipad mini
this would be a better experience on Android, the controller (gamesir x5) has a special Android mode where it can control features like the volume and apps get more access to things like the filesystem
this would be a better experience on Android, the controller (gamesir x5) has a special Android mode where it can control features like the volume and apps get more access to things like the filesystem
from the House Oversight Committee page
November 12, 2025 at 9:32 PM
from the House Oversight Committee page
the dump was shared as a public google drive, you can click around there.
Natives is a bunch of xlsx files and Text is a bad attempt at OCR or something
Natives is a bunch of xlsx files and Text is a bad attempt at OCR or something
November 12, 2025 at 9:31 PM
the dump was shared as a public google drive, you can click around there.
Natives is a bunch of xlsx files and Text is a bad attempt at OCR or something
Natives is a bunch of xlsx files and Text is a bad attempt at OCR or something
is it possible to run all the files through Deepseek-OCR and make them searchable through like Apache Lucene or something? (Like say q:”Summers” shows me the ocr’d texts and links to the images)
I don’t know the cost of doing so
I don’t know the cost of doing so
November 12, 2025 at 9:07 PM
is it possible to run all the files through Deepseek-OCR and make them searchable through like Apache Lucene or something? (Like say q:”Summers” shows me the ocr’d texts and links to the images)
I don’t know the cost of doing so
I don’t know the cost of doing so
I don’t need to read the King in Yellow, I played Signalis and watched True Detective season 1 I think I have the gist
j/k
j/k
November 12, 2025 at 9:03 PM
I don’t need to read the King in Yellow, I played Signalis and watched True Detective season 1 I think I have the gist
j/k
j/k
Larry Summers is reported in there, presumably Pinker isn’t too far off
November 12, 2025 at 9:02 PM
Larry Summers is reported in there, presumably Pinker isn’t too far off