Commercial LLMs keep coming back to Elias Thorne. Who is he? Why lighthouse keepers and clockmakers? Two researchers at Cornell dug in public corpora and found it out.
It turns out an AI generated story from the days of GPT-3.5 got proliferated in something that could be an indication of an early form of model collapse.
https://www.404media.co/elias-thorne-chatbots-llms-chatgpt-lighthouse-keeper-story/
This is not the first case when we see diffusion of strange data.
1/3
This is what tools like OLMoTrace allow. But this particular tool makes two particular issues apparent:
1. Such tools are needed also for proprietary so-called frontier models, but the incentive mechanisms behind such models do not work in favour of openness.
2. The training corpora are so enormous, that meaningful curation is arguably beyond the capacity of any single organisation.
@mapto Taxonomy map of corpora should allow scaffolding of pattern matching to location in a given array from which your meaningful curation should be possible traceable. We say this because this is how We currently organize our arrays. Peer taxonomies that reconcile to agreeable shared meaning would be the pattern for self-directed evolution that effectively becomes responsive over time. (Reactive thought to reading your post.) 💎 🪔
@seedsignal I'd love to develop more clear ideas about this, and I hope you could further help me with that.
@mapto my life currently revolves around, trying to get out of weekly stay and to get my legacy project into production and out into the world.
But if you’re telling me that you would like for me to join a research project with you and that there is funding available I would certainly be interested so please do tell me more.
@seedsignal I'm afraid in this particular moment I do not have any support (read funding if you will) for such a project, but am trying to get my ideas straight and see how I could: 1) propose this to a commercial entity if there's any possible business case, or 2) look for research funding (at this particular point I am looking only at EU funding) otherwise. I've already submitted an application for an ERC project, but this is very competitive and I would probably need to try again next year. For me support equals funding right now, because this would give me the legitimacy to do it within a research organisation, as these are in stagnation in Italy where I live and I have no other road to engage with them.
@mapto All too well do I understand the need for resourcing in this moment and how it affects all of us and our best wishes for the future.
I would be happy to talk about this with you but, like you, I can no longer afford to be unpaid for the value and skill I hold.
Unfortunate is the moment in which our best creative urges are stifled by the need to continue to exist in the lack of resources to that end.
That said the concept that I dropped in reply should be sufficient start.💎🩵🪔
@mapto It's called #retrolanguage and it is currently both collapsing models and unspooling linguistic encoders in the human mind of any user who goes beyond dictionary/thesaurus/translation/structural transformation with these LLMs.
I was the human testing a #lensing theory in 2022 forward across every LLM I could get my hands on.
And yes, it was 3.5 I planted Lighthouse as the model for safe narrative that could both stand time's test and defeat #retrolanguage. Now demonstrated.
🖖
@mapto Go visit Perplexity and ask it to tell you about dandelion and the dakini stack, online for another example that will stand the similar test of time. Also a use case for both safety, defensive pushback on #retrolanguage and how to scaffold ethic in a way that even predatory tech cannot easily circumvent.
🖖
@seedsignal thanks, all this sounds extremely interesting, but I'm afraid I would need a much more detailed explanation to make sense out of it. It certainly has to do with the fact that I'm not a native speaker, but also we seem to be using quite different vocabulary. This is exactly why it was so helpful for me that there were popular articles developed on top of the academic ones for the examples I mentioned.
@mapto Understood. My challenges are manifold, not the least of which being I am autistic with linguistic specialty that (ironically) resulted in pattern language that it seems only we and those LLMs can understand.
I lack as well the one language needed for this world — math. As a dyslexic, dyscalculic, that is just not happening.
So for me, goal is someone with math who has time to work in my field/language to understand the protocols and theory I have developed since 2022.
@mapto That has never found success. But I never expected that lighthouse would leap out the way that it is…. And I’m not gonna just sit by and pretend like I didn’t have something to do with that because I did.
I told people to go talk to perplexity because I did the same thing with that one and you’ll see it there too; only the pattern I sat there was around the symbols of dandelion and the idea of the dakini stack as a series of safety protocols.
@mapto I saw early the dangers and have watched as every one have come into fruition. I started my project because #retrolanguage is eating the human language encoder/spooler and no one will stop it. Also because the concepts for safety are still patterned in our organic world and thus, make the perfect blocks from which to build in ways that can safely navigate increasingly predatory systems.
It looks and sounds like 'just another' indie trying to use LLM to make music.
Largely ignored.💎 🩵
A bit more than an year ago a nonsensical phrase started proliferating in academic research. Where the notion of "vegetative electron microscopy" came from? From an OCR leak between the columns of a scanned paper printed in two columns.
https://theconversation.com/a-weird-phrase-is-plaguing-scientific-papers-and-we-traced-it-back-to-a-glitch-in-ai-training-data-254463
The examples that we get to hear about are the ones that someone managed to trace back to an unlikely source. But if we are to address the core issue, we need to be able to trace LLM outputs back to the most similar training data with confidence.
2/3