正在加载视频...

视频加载失败

launching our open source OCR tool today! try it out with some terrible pdfs and let me know how it goes:

342,717 次观看 • 1 年前 •via X (Twitter)

10 条评论

Karthik Kannan 的头像
Karthik Kannan1 年前

how are you dealing with hallucinations in the text if you're using GPT to OCR the text? while great for some use cases i've noticed hallucinations when it comes to dense text at scale. is there some benchmarks you're comparing against

Tyler Maran 的头像
Tyler Maran1 年前

hey Karthik. we're working pretty hard on a benchmark right now. Noticed the same issues with dense text (sideway text too). It's one of the big steps we need to take as we start on a fine tuning dataset. I'll be sharing as we make progress!

Bryan Lee 的头像
Bryan Lee1 年前

Waited a few minutes and nothing happened.

Tyler Maran 的头像
Tyler Maran1 年前

hey Bryan! sorry got more traffic than we expected and we had to up our limits. should be good to go now!

Dan Manastireanu 的头像
Dan Manastireanu1 年前

You should check out General OCR Theory, a specialized llm for OCR 2.0 Unfortunately, they don't have a gpt-4o benchmark for comparisson, but the results are pretty impressive

Daniel Tenner 的头像
Daniel Tenner1 年前

@ycombinator I’m literally just about to implement structured document OCR for accounts, tax computations and other similar docs, for my biz’s internal CRM. I thought “oh great I’ll use this instead of Textract”. But I can’t because of the “book a demo” thing. And no pricing. 😕

Harshil Prajapati 的头像
Harshil Prajapati1 年前

@ycombinator Make it open source

Tyler Maran 的头像
Tyler Maran1 年前

@ycombinator ... it is open source 👆

Tom Osman 🐦‍⬛ 的头像
Tom Osman 🐦‍⬛1 年前

trying now with some different one's but doesn't seem to be working? Using Arc if that helps at all.

Tyler Maran 的头像
Tyler Maran1 年前

we got a lot of traffic and had to up our limits! should be good to go now!

相关视频