Programming

As an experiment / as a bit of a gag, I tried using Claude 3.7 Sonnet with Cline to write some simple cryptography code in Rust - use ECDHE to establish an ephemeral symmetric key, and then use AES256-GCM (with a counter in the nonce) to encrypt packets from client->server and server->client, using off-the-shelf RustCrypto libraries.

It got the interface right, but it got some details really wrong:

• It stored way more information than it needed in the structure tracking state, some of it very sensitive.
• It repeatedly converted back and forth between byte arrays and the proper types unnecessarily - reducing type safety and making things slower.
• Instead of using type safe enums it defined integer constants for no good reason.
• It logged information about failures as variable length strings, creating a possible timing side channel attack.
• Despite having a 96 bit nonce to work with (-1 bit to identify client->server and server->client), it used a 32 bit integer to represent the sequence number.
• And it "helpfully" used wrapping_add to increment the 32 sequence number! For those who don't know much Rust and/or much cryptography: the golden rule of using ciphers like GCM is that you must never ever re-use the same nonce for the same key (otherwise you leak the XOR of the two messages). wrapping_add explicitly means when you get up to the maximum number (and remember, it's only 32 bits, so there's only about 4.3 billion numbers) it silently wraps back to 0. The secure implementation would be to explicitly fail if you go past the maximum size for the integer before attempting to encrypt / decrypt - and the smart choice would be to use at least 64 bits.
• It also rolled its own bespoke hash-based key extension function instead of using HKDF (which was available right there in the library, and callable with far less code than it generated).

To be fair, I didn't really expect it to work well. Some kind of security auditor agent that does a pass over all the output might be able to find some of the issues, and pass it back to another agent to correct - which could make vibe coding more secure (to be proven).

But right now, I'd not put "vibe coded" output into production without someone going over it manually with a fine-toothed comb looking for security and stability issues.

We should let these twits enjoy their shit on twitter. The AI hype is just like the crypto hype, it'll fade.

The name vibe coding sounds like a drunk evening with friends getting an MVP off the ground, but nothing more.

Anti Commercial-AI license

For personal projects, I don't really care what you do. If someone who doesn't know how to write a line of code asks an LLM to generate a simple program for them to use on their own, that doesn't really bother me. Just don't ask me to look at the code, and definitely don't ask me to use the tool.

So you mean debugging then?

That's a bad vibe if I've ever seen one.

Seems like a recipe for subtle bugs and unmaintainable systems. Also those Eloi from the time machine, where they don't know how anything works anymore.

Management is probably salivating at the idea of firing all those expensive engineers that tell them stuff like "you can't draw three red lines all perpendicular in yellow ink"

I'm also reminded of that ai-for-music guy that was like "No one likes making art!". Soulless husk.

They can vibe as much as they want, but don’t ever ask me to touch the mess they create.

Nearly every time I ask ChatGPT a question about a well established tech stack, it's responses are erroneous to the point of being useless. It frequently provides examples using fabricated, non-existent functionality and the code samples are awful.

What's the point in getting AI to write code that I'm just going to have to completely rewrite?

This seems like a game you'd do with other programmers, lol.

I can understand using AI to write some potentially verbose or syntactically hell lines to save time and headaches.

The whole coding process? No. 😭

fake

If you don't write a single line then you aren't coding

If it wasn't for the fact that even an AI trained on only factually correct data can conflagrate those data points into entirely novel data that may no longer be factually accurate, I wouldn't mind the use of AI tools for this or much of anything.

But they can literally just combine everything they know to create something that appears normal and correct, while being absolutely fucked. I feel like using AI to generate code would just give you more work and waste time, because you'll still need to fucking verify that it didn't just output a bunch of unusable bullshit.

Relying on these things is absolutely stupid.

Based on my experience of AI coding I think this will only work for simple/common tasks, like writing a Python script download a CSV file and convert it to JSON.

As soon as you get anywhere that isn't all over the internet it starts to bullshit.

But if you're working in a domain it's decent at, why not? I found in those cases fixing the AI's mistakes can be faster than writing it myself. Actually often I find it useful for helping me decide how I want to write code because the AI does something dumb, and I go "no I obviously don't want it like that"...

I probably wouldn’t do it. I do have AI help at times, but it is more for bouncing ideas off of, and occasionally it’ll mention a library or tech stack I haven’t heard of that allegedly accomplishes what I’m looking to do. Then I go research the library or tech stack and determine if there is value.

This sounds like something I put on my resume to get a coding job, but I'm not actually a coder.

It'd work, too.

Nah. I only used AI as a last resort, and in my case, it has worked out. I cannot see myself using AI for codes again.