Folks: Let's get Claude Code native installer working for FreeBSD

Uhhh... so, Claude Code is basically a shell script that interacts with the Claude service over HTTP. Not only that, the Claude service (the cloud end) keeps track of how you use it, and charges you a pretty penny for a subscription. How a security hole like that made it into FreeBSD ports misc/claude-code is beyond me. You'd think the committers have a better sense for what's a security hole and what's not.
:rolleyes:
 
And it is still lure-in prices, they're still supported by state grants and whatnot.
I will gladly use them for sporadic stuff in their online chat form as long as they're free.

I have a hunch that people who overly rely on agent-assisted 'coding' are going to receive a cold shower soon.
 
drhowarddrfine said:
So does it just stop when you run out till you pay up?
Click to expand...

In a meeting yesterday I heard from somebody who had to wait a while every time he hit a limit.

They can also downgrade you to cheaper models. And restrict you during rush hours so that users of more expensive plans come first. Generally there seems to be no hard limit, they seem to continue giving you something as long as it is convenient for them.

Note that none of the AI companies tell you specifically what your exact limits are (apart from the problem of being unable to predict how many tokens a given task might take).

People also expect an explosion of monthly base fees pretty soon.
 
mer said:
Why can't the AI figure out how to do a port of itself?
Click to expand...
Be careful what you wish for, openClaw agents may read your comment hunting for something to contribute to and then the flood of PRs to review will start on FreeBSD's repositories :) (if it hasn't yet). Just imagine what would happen if such an agent decided its mission was to create ports for "all software missing in FreeBSD".
 
My point was that a lot of software developers (including myself) are interested in using a given piece of software. Maybe they have had experiences, like I have, that make them believe this tool is helpful to them. The makers of that software have decided to put our OS, unlike OSX, Windows, and Linux into a "safe-to-ignore" category with e.g. Plan 9 or (for all I know) GNU/HURD. Since no one outside of our community will advocate for it, I'm contending that we should advocate for them to take FreeBSD seriously.

The objective debate about whether AI tooling is a net positive for a developer is largely irrelevant: clicky keyboards, vim/emacs, etc. As has been true togas were the fashion, to argue matters of taste is not to be done. As Kathy Sierra once said: when people use a bad tool, they say "this sucks;" when they use a great tool, they say "I rule and I use this." I'm hoping that folks who are looking to do something great look at FreeBSD, and when they find that it supports all their favorite tools, give it a try. Failure to furnish something to my base harness will disqualify a platform; and AI tooling is now in many folks' base harness.
 
sgharms said:
The objective debate about whether AI tooling is a net positive for a developer is largely irrelevant:
Click to expand...
the objective fact is that it is a net negative for everyone involved. even the slopbot companies own research backs this. but everyone who argues that it's good argues from the stance that it makes them, personally, individually, feel more productive, and what they tell us ("it really works!") is vastly different than what they show us (they spent an inordinate amount of time arguing with the chatbot, and the end result is indistinguishable from "look at this tasty corn i pulled from the sewage plant").
 
atax1a said:
the objective fact is that it is a net negative for everyone involved.
Click to expand...
We're talking very specifically about vibe coding, here, right? (or agents, or whatever term people use these days to say they use a LLM to do their work in their place). Because when it comes to using a LLM to *learn*, I can attest the result is in net profit big time. I every day paste errors or ask questions to LLMs and get quick directions about where I should look for, replacing the previous sometimes hours of reading of web pages, trying to figure what was even the proper search terms to use. And I'm talking here about tech, which is my job and expertise. When I ask questions about domains I learn as hobby and know nothing about (electronics, sewing, mechanical engineering, cooking, etc), the speed up in finding my answers are even more spectacular.

But yeah, it should have stopped there. Somehow, every time we get a cool new tech, you can count on the trend to be to push it to absurd extremes (OOP becoming hundred of layers of abstractions, service oriented architecture becoming microservices, containers becoming Kubernetes fleet and their "errors rate", dynamic web page becoming 10MB SPAs, etc).
 
you also didn't argue from anything other than how you feel when you get random text back from chatgpt. it's impossible to find things on the internet now because every query results in slop.
 
If you expect me to build an experimental protocol to prove my point to you, no it won't happen. If anything less than that is "just someone saying how they feel, so completely irrelevant", then there is no point in discussing on a forum.
 
objectively speaking, how you feel about the personal convenience benefits these things give to you is irrelevant, but, again, i guess people can excuse the massive externalities if the returned pap from chatgpt makes them "feel" more productive or whatever
 
When it takes me 30 seconds to find an answer instead 30 minutes, this has nothing to do with subjectivity. There is a serious subjectivity problem, though, when you try to dismiss such experiences from others to fit your own narrative using such sophism as "whatever experiences other people report are just their feeling". Then again, why do I even care. This discussion is going nowhere, I'm out.
 
drhowarddrfine said:
I hadn't thought about that. Was getting ready to give it a try soon so that's good to know.
Do they automatically bill you for that or do they just shut you down till you pay?
Click to expand...

So for Anthropic Claude it is like this:
- No API keys at all for the monthly plans. You pay extra for API key usage
- Monthly plans can only be used from provided interfaces, hence no API keys
- usage limits for plans are undocumented, flexible and soft
- API key usage is pay as you go by token, price depends on model Claude costs $15/milliontokens

The API is pre-paid so you cannot be billed more than what you put it.
 
LOL - When you "buy" something you usually know what you are paying for :cool:

This seems like putting out a lot of money up front and then reaching into the "mystery bag" to see what you get.
 
You must log in or register to reply here.
Back
Top