Hey there! For complicated reasons, some of which I will explain later, I'm in the middle of dong a pretty gigantic revamp to the Motte codebase. Right now, this means "fix a bunch of small bugs". Later, other things will be happening.
The immediate goal of this is not to massively transform the site or do a Digg v4/New Reddit thing. Frankly, if the site remained exactly the same (but with fewer bugs) I would be happy.
The long-term goal isn't going to be quite that static, though. The long-term goal is to set things up to allow more subcommunities, similar to how subreddits work. This will be a larger set of changes. These changes will not obsolete The Motte itself; I am absolutely aiming for a site that can comfortably keep holding The Motte. I've talked about this before if you want to go read about it - this is still the same plan, for what it's worth, it's just now being (slowly) executed.
I will be posting more, and asking questions about next steps, when actual site behavioral changes are in the queue. This just isn't that! This is infrastructure work.
tl;dr: The site backend may be a little wonky for a bit, sorry about that, please report any bugs.
(and mods, feel free to unsticky this in a few days if I forget)
Jump in the discussion.
No email address required.
Notes -
Don't tell me you're rewriting it in Rust?
It's probably some AI-coded slop shit. Sadly, the people who, now feeling empowered by AI, rush into updating a bunch of otherwise stable projects are, often, a bit too eager and lack taste and accept absolute horseshit for code. I hope the admins here have better judgment, but I've seen this happen too many times already to have any hope left.
If you can find bugs I will be happy to handle reports!
That said, I'm paying roughly no attention to the quality of the current code because it's about to get completely scrapped anyway.
More options
Context Copy link
I've come around somewhat on AI code and have come to accept that while the code is shit slop, it mostly works.
For one-off scripts that just need to work and won't go further, it's amazing. For small applications, it works ok. For larger projects or greenfield development with evolving requirements, it's pretty bad. The one constant of AI driven development though is that the final result will become thoroughly unreadable and unworkable for humans.
I expect that for a reddit clone scale service, it's probably workable, especially if the site doesn't blow up in popularity, but I would expect rough edges and weirdness. (not that human code is free of that either)
A quick oneoff script told me that 42% of the entire lines of code in the motte repo was created or edited in the last week. Make your own predictions on the outcomes of this.
Ironically, despite being wildly bullish on LLMs compared to you, I've become more skeptical of AI coding since moving to a team a couple months ago that's extremely AI coding focused (think dozens of people, each contributing many thousands of lines of code daily, all code reviews done via AI, as well as deployments, monitoring, design docs; a guy from OAI came in and is implementing his vision). Fairly sure we're headed toward disaster, though our "velocity" makes leadership very happy.
Yes of course, I still think a slop factory is destined to become a train wreck. But I do begrudgingly admit that small scale vibe coding projects that are driven by a single human can give usable results if the human driving dogfoods the results.
More options
Context Copy link
More options
Context Copy link
I would be quite interested in people actually making falsifiable predictions.
More options
Context Copy link
More options
Context Copy link
You can look at the code and judge its quality for yourself.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link