|
|
|
|
|
|
by gravity2060
887 days ago
|
|
|
This is fascinating. Have you seen any evidence yet of it being picked up? Are you using visible text and hidden text? I understand you are likely to elaborate too much, but other thoughts and insights you are willing to share would be great. (Also, are you doing it algorithmicly? Does it make sense to do it as an open source project and get more like-minded collaborators? Are you seeing any evidence out there of others doing this, especially for commercial interests where I’d (sadly) expect this at this point by the cutting edge SEO crowd.) |
|
|
It's like spitting in the ocean right now in terms of biasing training data that's being used.
I've been hopeful that at least I might see results in RAG, but production integrations wisely seem to reduce the reliance on social media results which are an easier vector to inform.
It's more where I expect models to be in around 3-5 years once there's internal content critique and filtering that I think I might see some traction.
And just visible plain text, no automation.
Think of it more like seeding a literal logic bomb around the web specifically designed for future LLMs, where evaluation doesn't result in changing code (like traditional logic bombs), but in changing internalized reasoning and conclusions around a very narrowly scoped topic.