Retr0id

An honest gap

Hm. The name collides with Nura, an audio company with a product called Nuraphone. I suppose any pronounceable 4-letter word has collisions.

slop

The main problem with macros (and similar language features) is that they turn every codebase into its own DSL. When done tastefully it makes the code more readable, but there's still an overhead for newcomers to the codebase - they need to learn your DSL before they can be productive.

LLMs may make this overhead less visible to you, but surely it's still there? I'd rather more of my tokens went towards solving the actual task at hand, vs figuring out a custom syntax (and re-learning it on every fresh context window).

The sandboxing was incompetent, but the broader problem is that imperfect sandboxing is an inevitability. Doing useful things with agents requires hooking them up to the outside world, in one way or another.

Ditto. I fear Opus 5 gave me a false sense of security of how easily I could spot it.

If someone makes a cheap copy of my unreleased game and it's bad, that's also bad for me, I think.

> Good devlogs don’t have that level of detail, to let people easily recreate games

The whole problem with LLMs is that you don't need the detail.

And it's not even a CC-like, it's a claw-like, afaict?

I thought maybe I just had an unlucky first prompt, but man, this thing is an absolute pain to work with. I'm getting innocuous bash commands flagged as "[Third-Party Attack]". This seems to be related to the command auto-approval, switching to manual seems to be helping.

Your premise is half right, and the half that's right is better than you think.

That's not really true.

Edit: wow that's a lot of downvotes! I'm surprised people can't identify a trivial reasoning failure.

> Opus 5.5 (1M context)'s safeguards flagged this session. You may be seeing this for the first time on an Opus model: Opus 5.5 (1M context) is more capable and has stronger safeguards as a result, which can sometimes flag non-cybersecurity work. We're improving these safeguards to reduce the amount of incorrectly flagged messages. Opus 4.8 is answering instead, or you can edit and retry with Opus 5.5 (1M context).

Yay, yet another model I can't use for anything interesting, even with CVP.

Switch 2 has real security features and a competent security architecture, and AI can't work miracles and find bugs where none exist.

AI makes mincemeat of something like Sony cameras, which only have security-through-obscurity. I started my project a week or so ago and I have all the decryption keys I could wish for, and bootrom code execution: https://github.com/DavidBuchanan314/ILCE-7M4-RE/ (Some of the code is hand-written, but only for my own edification)

I haven't even been using Ghidra, I just let the LLM use objdump and python capstone, and it figures it all out for itself.

Extremely well, aside from speed.

> Gzip scales linearly. I can gzip a one terabyte file.

In part because gzip only has a 32KiB window size, and I think it'd be at least quadratic within that window if you were going for optimal compression.

Are people really publishing blogs in docx format?

Anything you can do with ""real"" randomness can be done just as well with a seeded [CSP]RNG, including DoS prevention if you can keep the seed secret.

If you really want true RNG, you can inject a deterministic RNG at test-time and use a real one otherwise.

I've been pondering something I call "slop core, artisanal shell", as a way of keeping vibecoding under control. Slop core might sound like the thing you want to avoid, but as long as it's purely-functional (or perhaps, merely deterministic) it should be robustly testable. The "artisanal shell" keeps the thing human-understandable and human-modifiable, as long as you put some thought into the API boundaries.

So, why are there 3 of them? I can make several guesses but I don't see any concise answers on the page.

Isn't the appeal of those esp32 modules that they're already compliant, as a module? Or do you still need the product as a whole to be compliant?

In a poll of "do you want to be surveilled", "do you think mass surveillance is good", sure. But in terms of revealed preferences, people will happily give up privacy for mild convenience.

The median internet user doesn't even block ads (and associated surveillance). Facebook has 2bn DAU. And so on.

The widespread opposition to flock is unfortunately an aberration. The average person is largely indifferent to mass surveillance, and I think a rebrand has a good chance of working. They'll give it a try either way.

The bipartisan nature isn't that surprising, given that surveillance is more of an authoritarianism vs libertarianism issue i.e. orthogonal to left vs right (in an oversimplified kinda way).

A torrent with a webseed is strictly more resilient than a direct download link alone.

In regular unicode, a grapheme can be made up of an arbitrary number of codepoints (and thus an arbitrary number of bytes), which does cause issues at times.

True, I forgot about that! It's more fun when the inputs are printable ascii, though :P

It should also be easy to beat with just a few GPU weeks.

If anyone else has spare GPU hours they don't know what to do with, may I recommend partial hash collisions: https://github.com/DavidBuchanan314/birthday_party

(The "record" set by me only took about 1 GPU day - easy to beat!)

Stock photos and AI images might as well be in the same category tbh.