Asha Logos

This experiment will be my new permanently pinned tweet, and the only time I'll request retweets. The concept is simple: Post anything you find especially beautiful/inspiring - images, words, video, music - as often as you wish. Lets build a long eclectic thread of good things.

@civicheathen that can't be completely prevented and controlled, so it'll just be ignored.. I wouldn't be the biggest fan of it, but if it's helping get actual words of actual (esp pre-ww2) histories and scholarship out there, it could be worse.

@AshaLogos Are you worried that it will become a tool for people to copy+paste to grift and make quick money? Though I suppose it would just counterbalance the people already doing it with ChatGPT

Ok.. the 'Master Library' is created, and fully functioning - and thus the most obsessive portion of my project is complete, and I'll be back to active posting here and on SS and YT again. It was quite a lot of work, in all.. tens of thousands of historical books have been collected and included so far - many of which were older hard-to-find works, and likely would've disappeared forever, otherwise. Each night, as I sleep, the library automatically grows ever further - and two local LLM models further clean and organize and label the data. It'll only grow better, with time. I've worked with Claude Opus (and Gemini and GPT, but mainly Opus) to create a front-end to query this Master Library - lightning-fast searches, but I've also created an ultra-exhaustive overnight mode which spurs a team/council of local LLMs to answer the biggest and most important questions in the most comprehensive depth. The results are beautiful - as I mention in the article, I've learned more of historical substance in the past month or so from test queries of this Master Library than I have at any other period in my life.. virtually every historical work of note ever authored, down to the most obscure and hard to find, able to be queried in mere seconds. I can't imagine a more effective historical research tool - it's even better than I'd hoped and imagined. I'm not certain yet how I'll make this public-facing - I think the best idea, initially, may be an X account, in which people can ask it questions directly? I'm open to ideas... much more to come, hopefully soon - read article for comprehensive details.

@jackmaceoch agreed.. the ultimate goal is indeed creating a model from the ground up.

@AshaLogos Ultimately you will have to retrain from scratch a model if you want truth to be presented from the point of view of a time period that excludes the modern era. It is essentially embedded in the model weights of any trained on it.

@LuxLupus_ Yes, this is definitely possible - it just takes some time, and some doing. The end result would be something incredibly powerful and useful.

@AshaLogos You can you make it downloadable so that it can be run locally so people don't have to worry about it getting censored? I'd be willing to pay for a local version if it's possible.

A heavy preference for English at the moment, but definitely contains a wide variety, including much from the East

@AshaLogos Also, is it loaded with information purely from the Western Canon or also Vedic, Buddhist, Eastern texts, etc?

it's an excellent idea.. one I think I'd like to push as far as we can. If there aren't the most powerful and annoying/impassable legal hurdles, it'd be beautiful to find some way to create physical copies of nearly all (though perhaps prioritized according to demand/popularity)

@AshaLogos Thoughts on partnering with someone to release print copies of some works?

would be happy to share.. respect, for what you're engaging in. I fear that many of these may be impossible to find. I'm curious as to whether we might someday find some means of simply creating physical forms of all those for which we can't find existing copies?

@AshaLogos Asha I have thousands of these already in physical form, soon to be located in nice estate on some nice land in the states. If you send me the list I will slowly but surely track down all the remaining physical copies. Been at this for years. Let’s make the digital physical.

very tricky, and only very recently has success been possible imho - a two-stage process for the most difficult works like these, in which tesseract usually fails miserably but is followed by a very smart visual model, and the two are compared to find the middle-ground where there are serious doubts. A smart LLM can be trained to interpret the most tricky font types and archaic styles now.. likely too smart for most to run locally, but on a rented server it's certainly doable.

@AshaLogos @unheard_of_ How did you manage to get OCR to work with pre 1770s text? This is roughly the frontier where I’ve struggled to get it working well. The font type and set plus the quality of image of books before this was problematic

valid concern.. I have some built-in verification steps - if text is X% less than the original after any cleaning/etc step, it has to pass a secondary verification to confirm the truncation was necessary.. they do indeed hallucinate - or worse.

@AshaLogos I know it’s not possible for a human to double check every word of every book, but I’d just be careful about the AI truncating passages of books it’s been trained to disagree with. I remember Devon Stack having trouble with that a little while ago

root PDF files, many TB.. though many of these need OCR, and have various issues - but standardized format is markdown, much slimmer. It'd be possible to distribute this fairly easily.

@AshaLogos You should release a torrent of the material just in case. How many tb is it currently?

@Name03813069 I have so much ground to make up.. absolutely.

@AshaLogos Glad to see you back Asha you have been sorely missed! Video soon….?

@BasedTorba we should DM soon, my friend.. at the very least I'd like to know what you've learned on this front - and perhaps share a bit of what I have. Your efforts are much appreciated..

@AshaLogos As always, we are on the same page.

Epstein isn't a 'R' or 'D' thing, Epstein is an lsraeI and int'l finance thing - and a demonic and parasitic thing.

The only thing more disgusting than those making Epstein a mere political football are the handful of clowns who think the utterly hollow lsraeIi-intel pedo operative with the Woody Allen voice and a total lack of charisma has 'aura'... these are embarrassments and enemies.

@treblewoe @BasedTorba I'd like to talk with him - he's the only one who has done anything significant on this front thus far, and he's good people. @basedtorba

@AshaLogos this sounds like something @BasedTorba might be interested in helping with.

An account of the journey, and a big idea I'd like you all to consider: open.substack.com/pub/ashalogos/…

@rutgerkipling @p0lar_fawn it means he's terribly lost and disoriented.. as does anthropomorphizing God himself, believing the Creator of the universe might be capable of human-tier 'arrogance'. It's absurd beyond words.

@AshaLogos @p0lar_fawn Considering the opposite: what does it mean when a man refuses to praise something worthy of praise?

confession: maybe this is spiritually immature of me but i still don’t fully understand why God would expect worship. love, trust, obedience i’m ok with. but something in me resists the idea of God “needing” praise, even though i know that framing is probably wrong.. just trying to understand it without flattening God into a human ego problem

@PaulJustin27306 @p0lar_fawn We've always worshipped an All-Father - the furthest possible thing from Jewish, or poison.

@AshaLogos @p0lar_fawn Christianity is jewish poison.

@AgresvProgresv we should probably be in touch.. I've had to code nearly all of 'my' own tools, using claude/gemini/deepseek etc.. has worked well thus far, but I don't have the years of experience you do - I only know enough to guide the ship.

I used to work on creating AI tools for data augmentation and retrieval before I got fired for standing my ground on ethical issues. If you need a small team I’d be more than happy to lend help if you’re working in Python and are willing to share your codebase. Storing Data in a database and teaching AI to retrieve it made me a lot of money. It’s not nearly as easy as it sounds. I created a tool named TALOS to do just this. Happy to help.

I've spent the last few weeks quietly collecting every history book (especially those authored prior to the WW2 academia-takeover period in which this hypersensitive brainrot began) across the web, including OCR-ing those I couldn't find in text format, and organizing them cleanly into a library. I then had a 'LLM' (local language model) label each with a summary and tags, clean up title and author, and assign a few custom designed 'ranks' to each work - and then I indexed them all in what's called a 'RAG' (retrieval-augmented generation) setup, an ultra-efficient method of data storage/retrieval. This will be an ongoing process.. I'm gradually seeking out every single (significant or obscure) work ever penned, to create a permanent library.. I'm up to well over 10k relevant works thus far, a significant chunk of what's out there. Long story very short: I can now query this ENTIRE library simultaneously, seeking keywords or ideas or concepts, using several powerful downloaded and locally running large language models, and answer virtually any historical question imaginable, in a mere minute or two - with hundreds of accurate sources, quotes, citations. I can even have it write entire articles or arguments containing all of the most relevant returned quotes and context, to aid in my writing and video production.. but the ultimate goals are much bigger. During the process, having completed the first smaller library, my machine was compromised, and I lost virtually everything.. but that's another topic, and a story for another time. Anyhow.. much more to come - and a short video, describing what this may eventually and gradually turn into, and some of what I've learned along the way. I hope to be a bit less silent in the near future! Merry Solstice, Christmas, Yule, and so forth..