Kawin Ethayarajh argues cheaper serving of smaller models will enable highly personalized, style-mimicking AI writing
Dominant models prioritize broad appeal, making current outputs generic.
#449Kawin Ethayarajh@ETHAYARAJHThe bad mouthfeel of AI-assisted writing is downstream of everyone using 1 or 2 two models to write, and those models thus having to appeal to the lowest common denominator.
This is not the future. I've seen no convincing argument that LLMs are fundamentally incapable of imitating one's personal style of writing, only that such models are inefficient to serve. As smaller models get better and better + models get cheaper to serve (both locally and in hybrid setups), we'll reach a point where it's almost impossible to distinguish between human and AI writing. It will still be important to write things yourself, for the sake of your own thinking, but the signaling benefits will disappear.
What we need is a more robust sense of what is and isn't slop, whether it's written by humans or machines.
Dwarkesh Patel@dwarkesh_spPeople complain about AI writing, and I sympathize. Yet I often find it a net improvement to ask an AI to explain something I’m reading more clearly.