Some meta thoughts on the general idea - not meant to denigrate this tool at all.
The problem I see here is that you will never learn to master the command line tools themselves. This may make sense for someone who rarely uses the terminal interface and does not need to feel comfortable in that environment. For the rest of us, professional admins or old-school hackers, it is not a viable alternative. They need the reference of a man page to get unambiguous clarity about a command or option, but also for discovery and exploration. In practice, a command that has already been used will be retrieved via your shell history. We do not want to repeat ourselves and we do not want to write verbose prose, IMHO that is the opposite of the hacker spirit. More importantly, we want to understand what is going on with each command, especially when it comes to long command pipelines.
In general, I wonder more and more if we are all losing knowledge about the lower levels of computer-machine interaction. I mean, going up the ladder from assembly to higher programming languages and now to natural language reaches its tipping point where you lose the deterministic chain of control. You can always tell what a computer program is doing, but you can never be 100% sure of what a human is trying to express with his words and whether his actions follow. I am afraid that the same feeling is now crossing over into the world of computation by using LLMs instead of unambiguous programming languages.
For any not-too-security-obsessed readers of HN: this sends all of your shell commands to some server (ha, but you can choose which one!). We all make mistakes, we all once in a while paste a password in a shell by mistake, or private API endpoint, or you know, expose architecture of your system by history of your shell. Why would you use it?
Edit: Actually flagged it. I believe it brings more harm than help. Whoever is going to use it, they will not be able to assess impact. Whoever can assess the impact, they will not use it.
9 comments
[ 1.7 ms ] story [ 27.9 ms ] threadIt seems fine, but I would appreciate a nixos package for this.
on the other hand, How is the LLM configured? Does it support llama etc...? Does it provide an automatic way to install the LLM?
The problem I see here is that you will never learn to master the command line tools themselves. This may make sense for someone who rarely uses the terminal interface and does not need to feel comfortable in that environment. For the rest of us, professional admins or old-school hackers, it is not a viable alternative. They need the reference of a man page to get unambiguous clarity about a command or option, but also for discovery and exploration. In practice, a command that has already been used will be retrieved via your shell history. We do not want to repeat ourselves and we do not want to write verbose prose, IMHO that is the opposite of the hacker spirit. More importantly, we want to understand what is going on with each command, especially when it comes to long command pipelines.
In general, I wonder more and more if we are all losing knowledge about the lower levels of computer-machine interaction. I mean, going up the ladder from assembly to higher programming languages and now to natural language reaches its tipping point where you lose the deterministic chain of control. You can always tell what a computer program is doing, but you can never be 100% sure of what a human is trying to express with his words and whether his actions follow. I am afraid that the same feeling is now crossing over into the world of computation by using LLMs instead of unambiguous programming languages.
For any not-too-security-obsessed readers of HN: this sends all of your shell commands to some server (ha, but you can choose which one!). We all make mistakes, we all once in a while paste a password in a shell by mistake, or private API endpoint, or you know, expose architecture of your system by history of your shell. Why would you use it?
Edit: Actually flagged it. I believe it brings more harm than help. Whoever is going to use it, they will not be able to assess impact. Whoever can assess the impact, they will not use it.