you are viewing a single comment's thread
view the rest of the comments
[–] 49 points 11 months ago* (last edited 11 months ago) (15 children)

There's a 'meme' trend of local ML tinkerers messing with the Epstein files as a dataset: https://huggingface.co/datasets/tensonaut/EPSTEIN_FILES_20K/

See: text embeddings https://huggingface.co/datasets/svetfm/epstein-files-nov11-25-house-post-ocr-embeddings


Edit: Now I’m pondering making an “EpsteinGPT” finetune myself. Maybe like a 4B-14B model for the sole purpose of Epstein RAG? Or a 32B responding in the style of the Epstein email text, just because.

  • source
  • hideshow 15 child comments
  • [–] 29 points 11 months ago (5 children)

    Just imagine having to explain being in possession of a handmade "EpsteinGPT" to someone 🤣🤣

  • source
  • parent
  • hideshow 5 child comments
  • [–] 6 points 11 months ago* (4 children)

    Meme finetunes are nothing new.

    As an example, there are DPO datasets with positive/negative examples intended to train LLMs to respond politely and helpfully (as opposed to the negative response). There are some that include toxic comments plucked from the web as negative examples.

    And the immediate community thought was "...What if I reversed them?"

  • source
  • parent
  • hideshow 4 child comments
  • [–] 6 points 11 months ago* (2 children)

    haha just imaging people showing off their collections, "here's my Mr. Rogers chatbot, and Thomas Jefferson, and even Luffy from One Piece! And uh...oh yeah over here we have EpsteinGPT for when I, I mean for if, um...its for lulz ok?! Don't look at me like that, where are you going?!"

  • source
  • parent
  • hideshow 2 child comments
  • [–] 11 points 11 months ago* (last edited 11 months ago) (1 child)

    It's literally "this one is my fursona. This one won't refuse BDSM, but its not as eloquent. Oh, this one is lobotimized but really creative." I kid you not. Here is an example, and note that is one of 115 uploads from one account:

    https://huggingface.co/Mawdistical/RAWMAW-70B?not-for-all-audiences=true

    And I love that madness. It feels like the old internet. In fact, furries and horny roleplayers have made some good code contributions to the space.


    Early on, there were a few 'character' finetunes or more generic ones like 'talk like a pirate' or 'talk only in emojiis.' But as local models got more advanced, they got so good at adopting personas that the finetuning focused more on writing 'style' and storytelling than emulating specific characters. For example, one trained specifically to stick to the role of a dungeonmaster: https://huggingface.co/LatitudeGames/Nova-70B-Llama-3.3

    Or this one, where you can look at the datasets and see the anime 'style' they're trying to massage in: https://huggingface.co/zerofata/GLM-4.5-Iceblink-106B-A12B

  • source
  • parent
  • hideshow 1 child comment
  • [–] [S] 16 points 11 months ago (3 children)

    We need more of this.

    Not data hidden in some file.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 5 points 11 months ago (2 children)

    Instead of em dashes, it's full of extra spaces before and after each period and extra period.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 11 months ago* (last edited 11 months ago) (1 child)

    I dunno what the 'writing style' would end up as. The bulk of the text seems to be formatted like this:

    ...
    10. Is Epstein cooperating with federal suit against Bear Stearns hedge fund managers Ralph Cioffi
    and Matthew Tannin? Will he testify in their cases?
    
    11. Mr Epstein was deposed on this week, on Thursday. Is it true that he answered almost every
    question by invoking his Fifth Amendment rights?
    
    12. Defense attorney Brad Evans has filed a motion to freeze Mr Epstein’s assets. Has Mr.
    Epstein moved his money from the US offshore or abroad, or does he intend to, in order to
    protect his assets from possible damage claims?
    
    13. What did Mr. Epstein do during his work release program while serving time. Reports have
    said he engaged in “scientific research.” If so, what was he researching?
    ...
    
    Response
    
    "That's because it isn't, and everyone here
    (apparently save one) is rational and objective enough
    to understand that. Physical phenomena, and
    phenomena in general, are
    
    ultimately perceptual in nature and subject to
    observational replication - that's why they call
    physics an empirical science. But consciousness is
    not.
    
    Consciousness cannot be objectively, replicably
    observed. Its putative physical correlates, including
    ...
    
    Bill Clinton identified in lawsuit against his former friend and
    pedophile Jeffrey Epstein who had 'regular' orgies at his Caribbean
    compound that the former president visited multiple times
    
    e The former president was friends with Jeffrey Epstein, a financier who was arrested
    in 2008 for soliciting underage prostitutes
    
    e Anew lawsuit has revealed how Clinton took multiple trips to Epstein's private island
    where he 'kept young women as sex slaves'
    
    e Clinton was also apparently friends with a woman who collected naked pictures of
    underage girls for Epstein to choose from
    
    e He hasn't cut ties with that woman, however, and invited her to Chelsea's wedding
    
    e Comes as friends now fear that if Hillary Clinton runs for president in 2016, all of
    their family's old scandals will be brought to the forefront
    
    e Epstein has a host of famous friends including Prince Andrew who stayed at his New
    
    York mansion AFTER his arrest
    By Daily Mail Reporter
    Published: 09:06 EST, 19 March 2014 | Updated: 21:10 EST, 5 January 2015
    

    I'd have to generate prompt/response wrappers too. But it would definitely bring up Trump and Clinton randomly, heh.

    ...There are automated metrics to rank English text by reading level, 'quality' and such. I guess it could be filtered to most 'interesting' emails and reformatted.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 11 months ago (1 child)

    This is interesting. Do you have any thoughts on why someone would want to utilize the epstein data for ML? Like, what's the point, in your opinion? Just lulz? Or, something else?

  • source
  • parent
  • hideshow 1 child comment
  • [–] 3 points 11 months ago*

    Lulz.

    It’s an interesting coding exercise, though. Trying to (for example) OCR all the documents, or generate a relations graph between the documents or concepts, is a great into to language modeling (which is not prompt engineering, like most seem to think).

    If you’re like a reporter or something, it’s also the obvious way to comb through the documents looking for clues to actually make headlines. I dunno what techniques they use at big outlets, though.

  • source
  • parent