Why won't you respect your audience enough to write with your own voice?

This is not my area, but what about Rust or Odin?

Rust is slow to compile, and not that productive when writing gameplay code (see https://loglog.games/blog/leaving-rust-gamedev/)

Odin... technically not a scripting language, and haven't used it that much. But if compile times are good enough and hot reloading works, might worth a try. Though I think manual memory management can be antithetical when churning out gameplay code quickly (or when you're working with game designers with minimal background in programming)

Thanks. I know that there are things that can be done to bring down the compile speed of Rust, like splitting a project into sub crates. But I agree with the post that you linked. Game dev, especially UI and game mechanics benefits from a high degree of polish. You just can't realistically get that with compile times in the 10s of seconds. The most polished game I've played, Balatro, was built using Lua and Love2D specifically because of the iteration speed that they made possible.

You just reminded me of Star Wars Roleplaying on Habbo Hotel when I was a preteen circa 2006. Habbo Hotel was basically an IRC and MUD with user-created hotel rooms. The Sith and Jedi orders were popular "rooms" in Habbo, and would often rise to the top and bring an influx of users– mostly teenagers.

But they were clearly gated, like a nightclub. There were those who were "in" the Jedi Temple or Sith academy, and those who were out. This creates a sort of exclusivity, and people would naturally want to join the Jedi Temple or Sith Academy. They'd have to agree to a set of rules written on a sticky note on the room's wall, and dress the part.

Sometimes all hell would break loose when a Sith Warrior or Inquisitor would come and attack the Jedi Temple. They were legally allowed to do this, as long as people were online to witness it.

It was really interesting how they battled each other with just text chat. They had specified rules where you'd have to type out five action lines (minimum of 5 words) after an attack in order to maime or kill someone. At any point in those five lines, the defender could use 2 lines to repel the attack. To add a complication, the actions had to be coherent with each other.

So it ended up being a lot of;

Sith: Jedi filth! Hiding in your temple, while the universe suffers. I will free them of you!

Jedi: Your oppression ends here, Sith!

Sith: swings lightsaber towards your body

Sith: it connects with your neck

Sith: it cuts through the muscle

Jedi: swings lightsaber at lightsaber (illegal: 4 words)

Jedi: it parries your lightsaber away

Sith: it connects through the bone (illegal: parried)

Of course, these would often devolve into long OOC (out of character) arguments, and this drama would bring in even more of a crowd. It was chaotic, and a whole bunch of fun.

It wasn't just RP fighting though. They would have all the Jedi initiates wear "robes" and hang out in rooms decorated like a Jedi temple. Then the Jedi Masters would conduct lessons and we'd have to memorise the Jedi code.

I still remember the Sith code. I had to memorise it and recite it at random times to the Sith Lords. When I graduated from that, I ended up teaching it myself to several acolytes.

"Peace is a lie, there is only passion. Through passion, I gain strength. Through strength, I gain power. Through power, I gain victory. Through victory, my chains are broken. The Force shall free me."

Pretty fun times.

I'm a fiction writer. The way you make AI better at writing prose is the same way you make a fiction writer better at writing prose...

You edit it manually.

That's it. There are no shortcuts to good prose, except perhaps paying a ghost writer.

I've been waiting for my lobste.rs invite for years. I joined the IRC for a few days, but being in Australia, rarely saw too much activity. I could chase someone down for an invite, but at that point I just don't care that much.

I was given my invite from a decent person who, I believe, basically has an "open door" policy. I'm happy to pay it forward and give you an invite if you want it. abetusk@mechaelephant.com

I'm just dipping my toes into AI. What prospective guardrails are being considered or used?

> What prospective guardrails are being considered or used?

The parent mentioned two: (1) feature flags, (2) rollbacks. You can add tests in CI, etc. YMMV. The LLM will do that if you ask it. But someone needs to spent time on it, test, etc. And most engineers don't get paid to ship tests / feature flags / rollbacks ... they get paid to test features.

"It's all the same, only the names have changed..." - JBJ

> What prospective guardrails are being considered or used?

The most important guardrail we are using at the moment is: "Make no mistake"

/s

Just to piggyback onto this comment; has anyone tried running multiple of these in conjunction? For example, having a Python script that has one of these orchestrate others, and offloads certain tasks to better/more powerful models, or even cloud models?

yes but then that defeats the purpose of 'local'

and if remaining local, the hardware required to run multiple poor models could be better spent running better models.

I have attempted to orchestrate using different models, loading and unloading, but the speed is not there and by the time mistakes are discovered considering the lack of quick iteration the results become worthless unless the task is trivial.

For executives looking to impress? Not really. Being able to rattle off perspective on a book, curated by someone with very high media literacy would signal the same level of media literacy to their audience.

An LLM may be able to synthesise results well each time, but there will be quite a difference between a synopsis written by an LLM and someone whose job it is to write synopses of books.

Huge difference in quality, and considering the clientele, they are willing to pay for that quality.

There really isn’t anymore. It can ape anyone’s style well, including insights and almost no one can reliably tell if something is AI or not.

It's a fascinating question. I took the GPP as "media literacy" as more of an elite culture shibboleth. Making the right references in this sort of elitist signalling process is more about showing alignment to your contemporaries. It is just as much making the right references and omitting other references.

Being an LLM that "knows a bit of everything" doesn't necessarily give you access to know the audience expectations in this sort of environment. They are layers of fashion and social context which almost intrinsically embodied as a fringe of temporal currency and connection, not necessarily available in any training corpus.

An LLM could be stuck in some imposter/savant moat here, always making last year's references or possibly over or under selling the current expectation.

It only does this if you have the apps in multiple groups. Dump them all in one group. Yeah it sucks that you can't configure each of them separately, but it works really reliably now.

Look I love Rust and hate Typescript. But if NPM didn't exist, wouldn't the attackers just hit the next most popular supply chain? Cargo isn't immune to this, as much as I love Rust and wish more shops used it.