I'm still mad there's no straightforward way to convert a PDF into semantic HTML. There's plenty of tools to convert it into HTML that looks the same with pages and such, but I just want the content.
Sounds like a good opportunity for a crowdfunded start up.
Would it work to convert it to a simpler intermediate format like rtf or txt, and then convert into html? Why html anyway, Isn't epub more appropriate?
I just hate two column paginated lay outs. Give me pageless single column text.
We absolutely should have more specialized LLMs. That being said we have dozens of tools that convert documents and data. Also any engineer worth a nickel should be able to whip something up in an hour or so for most cases.
But I want AI to convert my mp3s to Oggs and vice versa 😭😭😭
Not some stupid "conversion library" or whatever that is
Oooh do you think that would work on my 96kbps mp3s?
Same for music like suno. I don't need to remix and hallucitate new fusion music, I just need a really good way to effectively search/discover all music that already exists in one place.
Isn't that just Spotify?
When I give suno a prompt for "electrojazz metal microtonal polyrhythmic" I get a fusion remix that sounds good, but sometimes it sounds just like a plagiarized remix of something someone created before.
But search in Spotify will not give me good results if I search for that. It's just the search that needs to become better for discovery, because the song type already existed to train suno, it is not truly creative.
Which is funny, because it should be faster, easier and cheaper to create a good search and discovery engine instead of training and implementing brute-force AI. Especially because there is so much music that there has to be some sample left when you search, but it needs to be less rigid...
tldr: I want spotify to be better at finding songs from a prompt than suno is at making a new song plagiarized from unmentioned influences.
So I guess it's just only us Millennials that know how to convert a PDF properly, and we're just sandwiched in between boomers and gen Z finding the most ridiculous ways to try to accomplish that task.
As always, GenX just forgotten.
Who?
Gen X is a mixed bag when it comes to this but most I've come across are more like Boomers than Millenials on this.
Yeah, I didn't want to shit on you guys. Most of you are all right.
Somehow, millennials ended up being the only generation that at least kind of knows how to use computers.
Naah. There is plenty of Gen X, Y, Z who know and plenty of Millenials who dont.
Its just if you wanted to "do stuff with computers" you had to develop some understanding back then.
Today you can "do stuff" like gaming much easier out of the box. So not everyone who "does stuff" knows his way around.
In the office most colleagues of all generations just know how to do their specific things, mostly in MS Office products.
I was probably using computers when you were still in your mom's ovaries. My first one was an Apple ][+ in 1982.
This is why millenials, despite all else wrong with us, are the best generation. I asked a kid the other day if they knew what a directory was..... Crickets.
What's fucking wrong * is * that we're sandwiched between Boomers/GenX and Zoomers. Too young to be able to shove the Cold War brain rotters out of power, too old to convince the Zoomer incels to talk to a girl instead of listening to Andrew Tate.
You can't parse [X]HTML with LLM. Because HTML can't be parsed by LLM. LLM is not a tool that can be used to correctly parse HTML. As I have answered in HTML-and-regex questions here so many times before, the use of LLM will not allow you to consume HTML. LLM are a tool that is insufficiently sophisticated to understand the constructs employed by HTML. HTML is not a regular language and hence cannot be parsed by LLM. LLM queries are not equipped to break down HTML into its meaningful parts. so many times but it is not getting to me. Even enhanced irregular LLM as used by Perl are not up to the task of parsing HTML. You will never make me crack. HTML is a language of sufficient complexity that it cannot be parsed by LLM. Even Jon Skeet cannot parse HTML using LLM. Every time you attempt to parse HTML with LLM, the unholy child weeps the blood of virgins, and Russian hackers pwn your webapp. Parsing HTML with LLM summons tainted souls into the realm of the living. HTML and LLM go together like love, marriage, and ritual infanticide. The <center> cannot hold it is too late. The force of LLM and HTML together in the same conceptual space will destroy your mind like so much watery putty. If you parse HTML with LLM you are giving in to Them and their blasphemous ways which doom us all to inhuman toil for the One whose Name cannot be expressed in the Basic Multilingual Plane, he comes. HTML-plus-LLM will liquify the nerves of the sentient whilst you observe, your psyche withering in the onslaught of horror. LLM-based HTML parsers are the cancer that is killing StackOverflow it is too late it is too late we cannot be saved the transgression of a chi͡ld ensures LLM will consume all living tissue (except for HTML which it cannot, as previously prophesied) dear lord help us how can anyone survive this scourge using LLM to parse HTML has doomed humanity to an eternity of dread torture and security holes using LLM as a tool to process HTML establishes a breach between this world and the dread realm of c͒ͪo͛ͫrrupt entities (like SGML entities, but more corrupt) a mere glimpse of the world of LLM parsers for HTML will instantly transport a programmer's consciousness into a world of ceaseless screaming, he comes, the pestilent slithy LLM-infection will devour your HTML parser, application and existence for all time like Visual Basic only worse he comes he comes do not fight he com̡e̶s, ̕h̵is un̨ho͞ly radiańcé destro҉ying all enli̍̈́̂̈́ghtenment, HTML tags lea͠ki̧n͘g fr̶ǫm ̡yo͟ur eye͢s̸ ̛l̕ik͏e liquid pain, the song of re̸gular expression parsing will extinguish the voices of mortal man from the sphere I can see it can you see ̲͚̖͔̙î̩́t̲͎̩̱͔́̋̀ it is beautiful the final snuffing of the lies of Man ALL IS LOŚ͖̩͇̗̪̏̈́T ALL IS LOST the pon̷y he comes he c̶̮omes he comes the ichor permeates all MY FACE MY FACE ᵒh god no NO NOO̼OO NΘ stop the an*̶͑̾̾̅ͫ͏̙̤g͇̫͛͆̾ͫ̑͆l͖͉̗̩̳̟̍ͫͥͨe̠̅s ͎a̧͈͖r̽̾̈́͒͑e not rè̑ͧ̌aͨl̘̝̙̃ͤ͂̾̆ ZA̡͊͠͝LGΌ ISͮ̂҉̯͈͕̹̘̱ TO͇̹̺ͅƝ̴ȳ̳ TH̘Ë͖́̉ ͠P̯͍̭O̚N̐Y̡ H̸̡̪̯ͨ͊̽̅̾̎Ȩ̬̩̾͛ͪ̈́̀́͘ ̶̧̨̱̹̭̯ͧ̾ͬC̷̙̲̝͖ͭ̏ͥͮ͟Oͮ͏̮̪̝͍M̲̖͊̒ͪͩͬ̚̚͜Ȇ̴̟̟͙̞ͩ͌͝S̨̥̫͎̭ͯ̿̔̀ͅ
The pony always gets me
Oh, the timeless classic. Chef's kiss
Breathtaking. I wish i had more than one upvote.
Alright new copypasta drop!
If it works and it's stupid, then it's not stupid.
Starving yourself to lose weight works. It's also stupid.
It is still stupid, if it is incredibly inefficient and super wasteful of limited resources.
pdf to brainrot
I used LLM to format my markdown table (ಥ ͜ʖಥ)
This is a primary use for me. A couple times per day.
That’s part of what makes LLMs so popular with software engineers, they solve lots of trivial daily computer tasks.
At the cost of only one (1) inhabitable planet!
send me your data and i will parse it for you
it may take me a week to get back to you
How many tokens fit in your context window?
About tree fiddy
if java then like 9
Yes, there are LLMs for that, you literally just have to Google "llm parse PDF".
You could also use tesseract or any number of other solutions which probably work as well...
But an inexperienced kid is gonna act like an inexperienced kid
But an inexperienced kid is gonna act like an inexperienced kid
Let's give him access to government payroll!
It's using a shotgun instead of a drill.
Dangerously.
Doge aside, He's a kid that won prizes for writing ml models to parse greek letters from unreadable ancient ashen scrolls. Inexperienced? You sure?
Did he tho?
It's kind of easy to forget about or ignore any experience they might have if they're asking questions like that. Sure, maybe it was a brain fart from a panicked intern who's having orders barked at them from a powerful individual that they want to impress, but that doesn't make it any better, does it?
I guess it’s just a dumb tweet question then. In not directly in the field and even I’m up to date in latest model benchmarks for tasks I need to do.
I don’t know anything about this kid other than the tweet and Elon likes him.
Oh yea, I posted the meme to the wrong community
I mean he's not wrong.
Edit: it seems the joke that LLMs just take other people's data and regurgitates it in another format went over everyone's head 🥺
Using LLM for format conversion is like taking a picture of an electronic document, taking the card out of the camera and plugging it into a computer, printing the screenshots, taking those prints to a scanner with OCR, turning the result into an audio recording, and then dictating it too an army of 3 million monkeys with typewriters.
It's true. This person gets it.
Source: Am one of the monkeys.
Haha considering just how much irrelevant third-party training data you'd be looping into a format conversion, this metaphor really is spot-on.
Sounds very appropriate for a government operation
Im not so sure. I think this is more of a question about taking arbitrary, undefined, or highly variable unstructured data and transforming it into a close approximation for structured data.
Yes, the pipeline will include additional steps beyond "LLM do the thing", but there are plenty of tools that seek to do this with LLM assistance.
So...my process (which you just accurately described) could be replaced by an LLM, after all? Hooray! Monkey feed isn't too expensive, but a million mouths is still a million mouths.
