Hacker News Reader

Comment by t-3 | Hacker News Reader

Why are you comparing LLM's to a human brain? Software should integrate software when solving problems. It's completely reasonable to expect an LLM given a "count lines" problem to just pipe the text through wc -l.

View on HN

An LLM itself can't use wc. Coding agents like Claude Code or Cursor will call out to command line tools for this kind of problem when the LLM detects it.

Well, maybe not wc directly, but they have access to sandboxed Python environments. It must be trivial for an LLM to write the Python code that calculates this.

I don't understand why Gemini insists that it can count the lines itself, instead of falling back to its Python tool [1].

[1] https://github.com/elder-plinius/CL4R1T4S/blob/main/GOOGLE/G...

which they will do I'd imagine after being told they have access to a shell

Most LLMs have access to such tools. Well, maybe not a Unix shell, but something similar. This is from GPT 4.5's system prompt [1]:

    python

    When you send a message containing Python code to python, it
    will be executed in a stateful Jupyter notebook environment.
    python will respond with the output of the execution or time
    out after 60.0 seconds. The drive at '/mnt/data' can be used
    to save and persist user files. Internet access for this
    session is disabled. Do not make external web requests or API
    calls as they will fail.

[1] https://github.com/0xeb/TheBigPromptLibrary/blob/main/System...

Reply on news.ycombinator.com