It's really the only way that anyone will be able to make sense of 2M documents. There entirety of Wikipedia is only like 7M articles.
I also don't expect that DOJ lawyers are going to be able to redact all of the appropriate names (victims) with any degree of accuracy. But LLMs are made for tasks such as this.
ADVERTISEMENT
Correct, you train a model on this data specifically.
Replies to: "Correct, you train a model on this data specifically."
- Why are we not using AI for this? [LINK]
- iairishcheeks - 3:27pm 1/22/26
(9) [View All]
- It's too busy make memes of Trump-Beetlejuice roaming around Greenland. [NT] - LanceManion - 7:56am 1/23/26
- Depends. For classified files, you'd need a closed A.I. source. - jakers - 5:08pm 1/22/26
- Correct, you train a model on this data specifically. - iairishcheeks - 5:56pm 1/22/26
- there have to be a lot of variables there that could cause serious liability issues if mishandled. - und67 - 6:18pm 1/22/26
- Humans have already messed this up. [NT]
- iairishcheeks - 8:05pm 1/22/26
- so who trains the model? it would be a totally new model, n'est-ce pas? [NT]
- und67 - 9:30pm 1/22/26
- Gov IT with sufficient clearance. - iairishcheeks - 11:00pm 1/22/26
- Will AI excerpt out Bill Gates’ name as a professional courtesy? [NT]
- TakethetrainKnute - 9:24pm 1/22/26
- That would be considered a feature, not a bug. [NT] - iairishcheeks - 11:01pm 1/22/26
- so who trains the model? it would be a totally new model, n'est-ce pas? [NT]
- und67 - 9:30pm 1/22/26
- Humans have already messed this up. [NT]
- iairishcheeks - 8:05pm 1/22/26
- there have to be a lot of variables there that could cause serious liability issues if mishandled. - und67 - 6:18pm 1/22/26
- Correct, you train a model on this data specifically. - iairishcheeks - 5:56pm 1/22/26
Consent Management