Any idea how you can gauge (preferably empirically) strength of local structures in text corpuses that these LLMs gorge on?? I’m guessing an assumption you’ve made is - all common LLMs are trained on strong locally structured data which explains why hippity-hopping works
This meme made for an incredible thumbnail.
Haha thank you
Any idea how you can gauge (preferably empirically) strength of local structures in text corpuses that these LLMs gorge on?? I’m guessing an assumption you’ve made is - all common LLMs are trained on strong locally structured data which explains why hippity-hopping works