The Batch from Andrew Ng recently mentioned llms.txt. For me, it finally puts specific language and framework around the idea of intentionally designing some or all of your content online to connect with and direct llms.
Interesting thinking about what are the limits, trends, and what will change.
I think most people have figured out that adding a lot of data into a chat, then pinging against it to get specific answers is pretty standard functionality with llms.
Thinking about how that specifically works online is interesting.
A long structured xml file is one part of the framework. I’m thinking it would need to be automated, so will probably either be a standardized toggle option. Probably the ongoing nuanced changes maintained and updated by the main group doing that.
In terms of the creative side, instinct tells me voice nuances could be effected. Like, you could include in the llms.txt some kind of nuance to effect the way a multimodal ai speaks when it reads out your site content. This is hypothetical. I'm assuming you could script and trigger for multimodalities and geolocation.
If you want to read more about it here's the Batch:
https://info.deeplearning.ai/next-gen-models-show-limited-gains-real-time-video-generation-china-ai-chips-blocked-transformer-training-streamlined