Mastodawn

For the 1,000th time: "AI" does not have agency and cannot think and cannot act.

Chatbots cannot "evade safeguards" or "destroy things" or "ignore instructions".

They do literally only one thing and one thing only: string tokens together based on statistics of proximity of tokens in a data corpus.

If you attribute any deeper meaning to this, it's a sign of psychosis and you should absolutely never use chatbots, possibly you should even touch grass.

Ariell 1d ago

"op-sec? Yah, I got it.
It's next to the triple-sec, right?"

@whyrl The big problem I have with things like Claude Code (aside from the environmental and ethical ones) is that it usually works - which is very terrifying to anyone who is experienced with programming or just cares about correctness. Tech that usually works is tech that occasionally doesn't. If you had a phone that usually worked, you'd throw it out and get one that always works at the earliest available opportunity. LLM-generated code is seductive because it really does work enough to be functional, but secure, maintainable code is in the details. A huge portion of CVEs are from tiny mistakes that escape notice. How many of those tiny mistakes does Claude output? These risks can be mitigated by thorough and careful human review, but the nature of LLMs encourages moving very fast, leaving correctness in the dust. I have a feeling that there's going to be a lot of job security for white hat hackers in the coming years.

Ariell Mar 16

Whyrl Mar 16

Whoever thinks it's a good idea getting software developers (or anyone else, really) to feed natural language queries into a non-deterministic model to generate software is quite insane.

Ariell Mar 13

First draft of a character. Still some details in the air, but good enough for now.

Show thread

Ariell Mar 13

@darkgriffin
I've played a bit of the LBP games by them already, though it's been a while. I do remember them feeling extremely polished and usable, though I didn't do much with its creation tools. I always forget dreams and lbp were the same studio.
Neither of the 2 projects I'm working on will need 3d level design though, so I have time to figure things out before I need to build something like that. (1 is 2d, the other a 4x that generates rather than places its 'terrain')
I definitely need to learn toolbuilding in general still, though something less involved should probably come first. Dreams will be added to my list of reference games to check for those tools

Show thread

Ariell Mar 13

@darkgriffin I'm newish to development, I've been in IT my whole life, but proper education is new and my code is a mess. A functional mess, but still...
I'm hoping at least for some helpful ideas for toolbuilding, so I'll have to check out Dreams. I'm not entirely sure what I'll actually end up needing toolwise for what Im working on yet.

Show thread

Ariell Mar 12

@darkgriffin I always assumed Id need to slap together internal tools to do it myself when I get to that stage. Maybe someone will have a better suggestion~

Show thread

Ariell Mar 12

Don't get me wrong, there are discussions and nuance to be had in AI. But you won't find it in the corporate slop machines. The ones boiling the oceans, tanking economies, destroying personal computing, plagiarizing on a massive scale, and engaging in disgusting forms of social, political, and market manipulation aren't where the nuance is.
I'm not going to make exceptions or excuses for playing along with those just because some entitled [redacted] throws a tantrum. IDGaF if its for a small time whatever, or an open source project, or whatever other justification they want to try and push. If you're using the corporate LLMs by choice, you deserve to be shamed, and your project deserves to die.
It's not entitlement to want a functional society and livable planet, and the large data-center based LLMs are in direct opposition to those.

Ariell Mar 11

If you're using one of the big corporate LLMs, there is no nuance. Only a nonce.