3 ms·
> We need llms to be able to tap that not add the same functionality a layer above and MUCH less efficiently. Agents, tool-integrated reasoning, even chain of
by celdon25 7mo ago
> We need llms to be able to tap that not add the same functionality a layer above and MUCH less efficiently.
Agents, tool-integrated reasoning, even chain of thought (limited, for some math) can address this.
- RagnarD 7mo agoYou're both completely missing the point. It's important that an LLM be able to perform exact arithmetic reliably without a tool call. Of course the underlying hardware does so extremely rapidly, that's not the point.