Reaction
Does tool use mark a paradigm shift for LLMs?
One post argues that coding agents still work by predicting the next token, even when they run terminal commands. A reply argues that symbolic tools and harnesses around LLMs amount to a shift toward neurosymbolic AI.
TLDR
One poster says he underestimated how far LLMs would advance without a major paradigm shift. He argues that tool use is not a new paradigm: coding agents can run commands, but still do so by predicting the next token. In a reply, another user argues that the move from pure LLMs to systems with symbolic tools and harnesses is precisely the shift that led to the advance.
Combined views
7.1K
2 Sources, first seen 8h ago