Gil, The Shitposter Of Ages

Do you know the muffin man? I’m told he lives on Drury lane.

lol

lmao, even

Your –number one– most valuable resource is time. I know exactly how LLMs, and ML in general works. I am under no illusions of “this is the best thing ever”, or “we’re all gonna die to skynet”.

LLMs, and ML in general, have made it so that the average employee no longer spends most of their time doing mundane bureaucracy and documentation. This is objectively a good thing. Instead of spending 2-3 hours drafting and writing a report, you can feed the raw data into an LLM, have it spit out a report according to a pre-prepared boilerplate, and then review it. This task takes Claude on the free tier, maybe a minute; for the super data-heavy stuff. Its’ accuracy is also ~99.8% correct, and the human review fixes that ~0.2% in maybe a few minutes after that. We are down from 2-3hours of tedium, to maybe 4 minutes of double-checking. That’s a massive time-save.

Modern open-source models can be run on anyone’s hardware (at varying levels of speed), making the only benefit of datacenters, as mentioned above, time. If I can offload a task that would take 5 hours to run locally, onto a datacenter rack that completes it in mere seconds, that is a very good trade.

everyone who downvotes this post are armchair activists. pay them no heed.

LLMs are exceptionally useful as search engines and information collators; the recent daybreak models from OpenAI are actually impressive, they’ve been finding vulnerabilities in software thats gone unnoticed for years and even decades. the future is computer-assisted-intelligence, whether these armchair activists like it or not.

if your opinions are so vehemently in disagreement with mine, or any others’; then go out and do something about it. don’t cry and whine on the internet. write to your local congressmen/senators/prime ministers/etc. go out and sacrifice your comforts for your beliefs, but don’t -ever- say that “oh boohoo woe is me, the world is shit and all I can do is whine and cry on the internet”.

now with that being said, I personally have found Claude to be exceptionally useful in information collation; with local hardware running abliterated/uncensored/heretic models for information-secure stuff, so that tasks can be set up to run overnight. its been quite the useful tool.

edit: already got sixteen downvotes, LOL

Go back to your mommy’s basement and cry over a plate of nuggies and chips lmao

I’m already me. why would I want to be anyone else?

seems a bit conceited to me, to want to be something you’re not. leave that to the actors and the actresses, just be yourself. its perfectly okay to be yourself and nothing else… after all, it just takes some time, little girl, you’re in the middle of the ride. everything, everything, is gonna be alright (alright). above all else, just never let them distract you from that time in nineteen ninety eight when the undertaker threw mankind off hell in a cell and plummeted sixteen feet into the announcers table.

I don’t have an opinion on most things…

but I do have popcorn. hoo boy, there’s a lotta popcorn.

nah. its totally accurate.

there’s a lotta nutjobs here…

heck, its every one of us! we all had to come from our mom and dad at some point lolololol

same. I’m here for the memes, and the shitposts

I’ve not personally tested such stuff (as my setup is complete and I did the metadata all by hand before the age of LLMs), but to my understanding; the LLM is able to write opf/XML/whatever files you want for metadata, or even edit the file directly to add metadata. epubs and PDFs support author/title/etc tags iirc, so its not too hard to get the LLM to apply those after ingesting the text and searching online for matches.

I shitpost a lot in comments, making shitposts is lame and for losers. better to actually say meaningful things with a post rather than a nothingburger shitpost xD

I highly reccomend using calibre.

bonus points if you use a self hosted LLM to auto catalogue and apply metadata to the file before importing into calibre.

the recent openweight models are surprisingly good at reading files, pulling info, and doing simple monotonous tasks thousands of times without failing.

and thats where a self hosted llm comes in. feed it the files, have it apply the metadata, and bam you’ve got a library.

iirc calibre has a function like this now, its kinda cool IMO.