Remote
Cameron Dutro
@camertron@ruby.social
mastodon 4.7.3Ruby developer on the Core Product Dashboard team at Cisco Meraki.
212 Followers
234 Following
50 Posts
Joined June 10, 2023
"LLMs have democratized software development!"
No, LLMs have democratized dependency.
The frontier labs are using the same playbook as drug dealers. Give everyone cheap access to a replacement brain until individuals, businesses, and entire industries are dependent on it for their very existence, then jack up the prices and watch society burn.
Look, if you want to write some code with an LLM, fine. You shouldn't, but fine.
However you absolutely do not get to say *you* wrote it. You didn't. The computer wrote it for you.
If I buy a car from Honda, I don't get to say I made the car. It's the same principle.
We have to stop normalizing language like "I made x with Claude."
No you didn't. Better: "I paid Anthropic to write x for me."
The former is intellectual dishonesty.
Replying to
@jaredwhite@indieweb.social Agree! I'm at RubyConf right now and someone literally said they don't remember how to use vim anymore because they haven't written code by themselves in a year.
I refuse to get caught with my pants down when this all explodes.
Replying to
@jaredwhite@indieweb.social It's all about the speed of development, which I understand to some degree (I'm told cocaine also makes you go pretty fast 😏)
It is genuinely shocking tho how many folks are willing not only to endorse all the negative second-order effects of gen AI, but also happily sign away their own freedom.
It makes me feel like we learned nothing from the social media era.
I'm thrilled to share that I'll be speaking at @rubyconf@ruby.social in Las Vegas this July! 🎉 This year's program is exploring some genuinely exciting territory: community, craft, and the beautifully weird corners of Ruby. Can't wait to be part of the conversation. See you July 14–16! #RubyConf2026
Replying to
@jhwgh1968 it's wild how many times I've thought about leaving tech and driving a bus or something instead.
Replying to
@mark For sure. Computers exist to automate tasks humans would rather not do.
But it's a false equivalence to say LLMs are like compilers.
Compilers are deterministic and verifiable. The instructions produced by the compiler follow directly from the higher-level code.
LLMs are non-deterministic. The code they generate does not (at least, not formally or verifiably) follow from the prompt.
Replying to
@kerrick So why is it any different when Claude is involved?
It's not really a question of ownership either, since humans can own things that other people (or computers) created.
The US Copyright Office issued a report a few months ago that said AI-generated works cannot be copyrighted. So legally speaking it's not even possible to own the code Claude produces either.
I've come to the understanding that if you're not constantly second-guessing your parenting decisions, you're probably not a very good parent.
Replying to
@axol_otleon it's the "growth at all costs" mindset. Things working is incidental to the stock price going up.
Companies like Cloudflare mandate and incentivize AI usage, then cut staff when AI usage inevitably explodes.
If you didn't think AI was about jobs, think again.
Replying to
@arclight Yeah, that's a fantastic point. In my original post I was trying not to touch on the ethics of using LLMs since it's such a massive topic, but you're absolutely right.
There is no ethical way to use LLMs for the exact reason you mentioned - they are large-scale IP theft machines.
Replying to
@pointlessone One could argue that LLMs are really just a giant collection of pre-made configurations 😏
I also like the example of creating a Spotify playlist. You didn't sing or produce the songs, or work on the Spotify app. All you did was add songs to a list. So yes, you "created" it, but... not really.
I don't understand why everybody likes TOML so much. It's so inflexible... but maybe that's the point?
While Java's slogan is "write more, do less."
There is no such thing as being "ethical to a fault." Sticking to your ethics even when it doesn't benefit you is the whole point of ethics.
I just filled out Planet Argon's Rails community survey, and you should too!
Post about a cool OSS project it took me 4mo to make: zero engagement.
Post about speaking at RubyConf about a cool OSS project I'm working on: zero engagement.
Post a Calvin and Hobbes strip from 1995: Thousands of favorites and boosts.
At least now I know what the people want!
Replying to
@pointlessone@status.pointless.one Hahaha yeah, I meant "shit" like "stuff." We went to the park and I was carrying 2 scooters, 2 water bottles, a backpack with snacks/hand sanitizer, and 2 helmets 😂
Replying to
@gilesgoat lol or the one where he procrastinates because he thinks he and his future self are different people. "This is future Calvin's problem!"
Replying to
@jaredwhite Lol I would take a bad human programmer over an LLM any day because at least I can teach the human.
I tried to use Copilot to fix a segfault in a side project last night. It produced a working but inefficient solution, and I ended up having to fix the issue myself.
This happens like every single time I try to use one of these things. They're almost always net-negatives on my productivity.
Dunking on Anthropic is so fun and I am absolutely here for it.
But we would all be dunking on them even harder if they 𝘩𝘢𝘥𝘯'𝘵 used Claude Code to develop Claude Code 🤣
Normalize replacing "Claude" with "Clod" every time you see it.
Replying to
@lolitechengineer I'm actually glad you brought this up so I can refute it.
There's a big difference between a deterministic text transformation algorithm verifiable at every step (compiler) vs a pseudo-random non-deterministic next-most-likely token predictor (LLM).
The machine instructions a compiler produces follow directly from my code, so in a very real sense I did write them.
The code an LLM produces does not directly follow from the prompt, therefore I cannot claim authorship.
Replying to
@pointlessone@status.pointless.one It's more that parenting is really hard, so if you don't think you're making any mistakes, you're probably very wrong.
ngl it's so disappointing olive oil tastes nothing like olives
parent, /ˈper(ə)nt/, noun: Manager of all the shit
Replying to
@weekend_editor@mathstodon.xyz @joeldrapper@ruby.social That's a bit heavy. I'm not complicit, I just have like 1000 other things to worry about, and I just don't believe Ruby Central acted with malice.
Replying to
But seriously. How is this still happening to Windows machines in 2026? How does anyone still take Windows seriously? What absolute hot garbage.
Replying to
@soulcutter I think you are likely correct. One of my big problems with LLMs is that they may or may not be inserting vulnerabilities, but because nobody looks at or reasons about the code, we can't know for sure. However, since LLMs work by replicating the average human's output, and humans introduce vulnerabilities, then it follows that LLMs do too.
Replying to
@pointlessone@status.pointless.one Weirdly yes! It (well, a version of it anyway), was just made by someone I really respect 🤷🏼♂️
Replying to
@joeldrapper@ruby.social "Never attribute to malice that which is adequately explained by stupidity." -Hanlon's Razor
Replying to
@weekend_editor@mathstodon.xyz @joeldrapper@ruby.social Ok, maybe they acted with malice towards André, they seem to hate him. But not towards the other maintainers, IMHO.