There are a number of projects working on evals that can check how 'smart' a model is, but the methodology is tricky.
One would want to run the exact same prompt, every day, at different times of the day, but if the eval prompt(s) are complex, the frontier lab could have a 'meta-cognitive' layer that looks for repetitive prompts, and either:
a) feeds the model a pre-written output to give to the user
b) dumbs down output for that specific prompt
Both cases defeat the purpose in different ways, and make a consistent gauge difficult. And it would make sense for them to do that since you're 'wasting' compute compared to the new prompts others are writing.
I like how Loreline transpiles to multiple languages.
I wonder why a transpire to Inform 6 or Inform 7 has not already been done as it seems an easy conceptual leap. Perhaps I don’t fully understand the use cases here.
This seems more like Twine than Inform. Inform is for parser fic, where you enter commands ("get key", "go north", "open door"), Twine and this are for choice-based fiction.
It’s closer to Ink or Yarn Spinner (or to some extent Twine). Loreline isn’t doing any user text parsing, but is pretty good to manage branching dialogues and choices.
A theory that’s floating around is that since frontier models are so good at sounding like humans, companies paying for ads are arguing that Dead Internet Theory -> ad costs should go down.
Therefore, the push to ID everyone using the internet (even down to the hardware) is a way to prove that ads are being served to real humans in their target demographic.
It makes a lot of sense, too. Previously, governments wanted everyone to have to swipe their driver's license before accessing the internet. But now, businesses want it too. And that makes all the difference in a world built on capitalism.
If I had to guess it’s because GDP growth in industrialized countries are driven by financialization facilitated by cheap power and energy, and the countries that are more financialized don’t exactly want the countries with precious minerals to be industrialized enough to be self sufficient.
That’s why a lot of global south countries with valuable materials (oil, data center building blocks, etc) are often chronically destabilized, but not so much so that it’s impossible to find someone to buy unrefined materials from.
> All these “we” statements should be “me” statements.
And all of your sentences ought to have "I think that" at the beginning; while reading OPs message as a sign of victimhood instead of a statement of fact gives you the opportunity to sharpen your sense of self against it, it gives the appearance of signaling more than it gives actionable advice. otoh, if the former was your goal, I misread intent.
>If all you seek is comfort you’ll find it as you sit alone eating take out in a fully furnished apartment waiting to die. If what you want is community and to be present in the only life you have, you have to work hard to find it and maintain it.
Hard to argue with the last line though. Many people become victims of society through their own inaction.
The nuance that's lost in the article:
- Game Devs are similar and different to SWEs. The differences are that most game devs are contractors, work long hours, usually get underpaid given the work they're doing(they're working with low level languages to make droplets on Spiderman's suit look realistic enough for Youtube commenters).
- AI is 'good enough' for a lot of industries, and will be continued to go to market when it's 'good enough' for a given industry.
- That said, AI in games as it stands is really buggy. A "Good Enough" relationship, dialogue tree, Social Graph LLM for a game might take so much time finetuning that it may always be better to just start with Twine or an excel sheet.
- Gamers (with a capital G) are different from gamers. Gamers are the gaming equivalent of armchair food critics. They hate AI, and will brigade anything with a whiff of it, and will demonize anyone or anything that pushes back on them, even if the Gamers are wrong (see the 'Baldurs Gate Xalavier' drama).
- The issue is that Gamers are the influencers that determine whether the long tail of games (games that aren't AAA mainstays, yearly sports games, or the top multiplayer games) get enough momentum and traction to pop up for less serious gamers.
- Gamers hate AI, and will do everything in their power to make every content creator 'acknowledge the controversy' (these 'cancel culture'-adjacent dynamics are all downstream from Gamergate). Comnsidering Gamers are often the enthusiasts most game studios need to swing digital game store algos in their favor, you don't want to build a game that becomes a 'stand-in for a controversial topic.'
- If the game is polished enough and is a AAA game that has a dedicated audience, it'll get bought in spite of that.
- If the game was AA or indie, the influencers who'd amplify the game in other instances will talk about it like the end of True Artistic Gaming.
- This leads to less MTX-like games getting funded, because the long tail of gamers are mobile, and most mobile games are miniaturized one-way casinos with WoW guild warfare grafted onto it.
- If there are more layoffs as the gaming industry's financiers go risk off for lower ROI games, then only the AAA, P2W, gacha, battle pass games get funded, which creates a feedback loop.
So yes, games as an industry will continue to rake in money because gacha games, battle passes, and MTX business models make the overwhelming majority of the industry's revenue. What's being missed is that what saved gaming from the ET gaming market crash and took it to $1 billion in sales for Counter Strike lootboxes was the passion and specialized labor of a lot of people. A lot of issues with game development preservation have accumulated (losing source code, laying off people who knew how to build critically acclaimed games before doing knowledge transfer, etc), and will compound as thousands of people get laid off, and people either dial in core parts of games, or outsource logic to an LLM.
The article acts as a lamentation, because the games that most people above the age of 20 grew up with are not going to be made as often anymore, unless there's some way to introduce a whale/guppy power dynamic with online play, or a battle pass, or gacha "Pay-to-Win" mechanic that pays for itself within a few weeks of launch.
they could, but a law enforcement agent looking for a suspect will send a lot of subpoenas to every porn site. When a porn site says "we wipe that data instead of storing it," the law enforcement agent will say "what do you mean you wipe KYC and identity verification trails once you get them? Are you letting sanctioned people use your site and covering your tracks?"
Similar thing happened to Valve; people were trading gun skins, and regulators fined them for not having AML/KYC controls because the state argued "the business didn't do enough to stop money laundering."
This trickles out to porn companies (and the vendors that use them for identity verification), and implies that they need to store this data to prove that they didn't delete it to help terrorists.
The core axiom is unfalsifiable because it hinges on asserting the intent, and relying on other unfalsifiable questions to try and ascertain intent ("why do decentralization if not to avoid legal risk of doing bad things?"), but the answers proposed by crypto proponents aren't ideals or scenarios that the writer empathizes with, it seems
There are a number of projects working on evals that can check how 'smart' a model is, but the methodology is tricky.
One would want to run the exact same prompt, every day, at different times of the day, but if the eval prompt(s) are complex, the frontier lab could have a 'meta-cognitive' layer that looks for repetitive prompts, and either: a) feeds the model a pre-written output to give to the user b) dumbs down output for that specific prompt
Both cases defeat the purpose in different ways, and make a consistent gauge difficult. And it would make sense for them to do that since you're 'wasting' compute compared to the new prompts others are writing.
reply