I have to admit, PDF parsing being such a hot and profitable topic in computer science was really something I never saw coming.
PDFs? The things you can select text from? And when not, there's decent OCR? And when not, you just ask the person to send you an email or a word doc?
It sounds like LLMs are looking for a new unpolluted source of historical data that they can learn from, and this source exists in the form of old scanned-in paper documents. That's the only reason I can fathom as to why this is such a big thing now.
Every time I try to convert a PDF to epub or something, or OCR one that doesn't actually have selectable text, it turns out shit. I assume the real reason people would want to get LLMs involved is that there is actually a lot of ambiguity in what a correct conversion would be, and there are a lot of PDFs out there.
Training the most insane AI model on classified federal documents.
Ah yes, the famed document <-> JSON converter.
It's so easy! Watch:
{"contents": "<garbled .docx contents goes here>"}
This is that special blend of Tablet Kid "I don't need to know things I can google them" and Rich Kid "I don't need to do things I can crowdsource them" that makes for that Distinctively VP "I don't know what I'm doing and nobody can tell 👈😎👉"
Like opening source code in Word.
Technically OCR is an application of machine learning.
Not an LLM, though.
A world of difference
regularizing the OCR'd form into a json/html file might be a good application of an LLM though. Perhaps this is what they were asking about.
The secret to success in software engineering:
- Lie and say that there is
- Write or use a conversion algorithm
- Boss won't know the difference
- Collect bonus at performance evaluation
- Put "AI engineer" on resume
- Boss thinks AI can code at senior developer level and fires you and the entire team
- Never plan on staying at a SE job for longer than a few years. Not in a market that volitile.
No need, there's an unmaintained javascript library for that (written by a 12-yr old)
Initially, I didn't think these kids were fall guys.
Now I think they're fall guys.
That was my thought. Young kids fresh out of school are really easy to manipulate into delusions of grandeur, especially when said delusions are offered by the richest person in the world. He's gonna leave them out for the wolves.
Yes it’s an LLM called pandoc, you can run it locally
You don't need a private nuclear plant to run it? Wow very efficient.
Black magic software.
Learn more: openai.com/pandoc
Imagine getting a job like this and now half the nation knows your name...thats terrifying. being an intern may mean you have no idea of the true scope of what they are asking you to do.
They are public employees who are changing things at the core of our government. Why wouldn’t we know their names?
Government employees names aren’t secret (asides from a few exceptions) nor is their pay
We know that his dad is an engineering professor at university of Nebraska too. Really calls into question his credentials. I checked the other day and they had already removed his contact info from their website.
Oh yeah the hosted DeepSeek has that
Is this fake?
For context, this is the guy who figured out how to see what's written on some ancient Greek Scrolls without destroying them. It seems slightly far-fetched that he wouldn't know better.
What Greek scrolls?
Ok so they were apparently in Greek but not from Greece. Source: https://news.unl.edu/article-2
Actually this is what they do.
Yeah, I don’t really get this one. The class clown is the kid who recognizes a function of a tool, correctly at that. Unlike a dipshit lawyer who let it hallucinate bogus case law. Hilarious.
There are programs that exist that explicitly do these sorts of things.
They have been around since long before llm.
This is a lazy and uneducated question.
He demonstrates he has done zero research and goes straight to the buzzword because he knows nothing.
Just use Deepseek for US government data .... what could go wrong?
How is that a stupid question?
Large Language Models are for natural language processing, not for converting between text document file formats.
