Stop the AI Reporting Requirements

[GreenJelly]

As a software developer with 38 years of experience, I can attest EVERY game engine on the planet is NOW being maintained, and created with the help of Generative AI. AI is integrated in our IDE's, and we are using it EVERYWHERE. In addition, almost every game engine has AI built into it, and has had fuzzy logic for MANY years. Also, AI is being used in administration, data analysis, and many other area's. All of these usages are not required in disclosure.

In fact, I challenge you to find a single company not using Generative AI. Yet these reporting requirements only target Indie developers. This is leading to INDIE developers getting review bombed, and we are tired of it. STOP THIS. Remove these anti-AI review bombs, and remove these questions. Or instead, please check EVERY game on Steam as being built by AI.

Case in point: Microsoft is a key player in the AI development, yet all of their games are not marked as using AI. Go, look at a Microsoft game, is it AI marked?

2nd Case in Point: Lord of The Rings (the movie) used AI in the production of the movie. Their games, which also used scenes from the movie, also do not report generative AI usage. This is wrong, they openly admit to this, yet they don't disclose.

We're tired of the review bombs, the death threats, the angry competitors bombing us because we're honest and report usage, while they don't.

[Boblin the Goblin]

Początkowo opublikowane przez lx:

im not sure, but i think your translation is bad. surely its ambiguous

You never address the issue with something being open source for AI. You just stated "how much data is informative enough" when open source(which allows seeing everything) is what people want.

[Ben Lubar]

Początkowo opublikowane przez Boblin the Goblin:

Początkowo opublikowane przez lx:

im not sure, but i think your translation is bad. surely its ambiguous

You never address the issue with something being open source for AI. You just stated "how much data is informative enough" when open source(which allows seeing everything) is what people want.

Open source doesn't allow you to see where the data came from, it just allows you to see the data.

If I open source a book, that doesn't give you information about whether I plagiarized it.

[lx]

goblin, when you said open source, i found that sufficient and uninteresting. then i asked the strawman guy a question. then i dont know what you are talking about anymore.

users of those ai, are they not bound to give credit(disclose) by a license or agreement?

[Boblin the Goblin]

Początkowo opublikowane przez lx:

goblin, when you said open source, i found that sufficient and uninteresting. then i asked the strawman guy a question. then i dont know what you are talking about anymore.

users of those ai, are they not bound to give credit(disclose) by a license or agreement?

Nope.

[lx]

free lunch?

[MonkehMaster]

38 year developer, but person is only 37 years old?

that doesnt math right...

[Chika Ogiue]

Początkowo opublikowane przez Ben Lubar:

I've played many, many games that were translated into many, many languages, and they all used this technology called "a human" to translate them because it's cheaper than having to hire someone to go over all the wrong translations and fix them.

And that HUMAN was using a form of LLM because that's how we've translate these days. See, you've got it the wrong way round. It is cheaper to run machine translation -- which is essentially LLMs -- and then pay the PMTE rate which is LOWER than the translation rate. How low will depend on how the PMTE is being billed. If it's per source word (or character, depending on language) then that rate will be LOWER than the translator's translation rate. If it's paid by hour, it's significantly lower.

For example, a Japanese to English translator will typically charge between 6 to 12 JPY per source character to translate from Japanese to English. Their PMTE rate will typically be 4 to 10 JPY. Most agencies that hire for this have even lower rates for translators, as low as 1 JPY for translation and 0.80 JPY for PMTE. The shift from translation to PMTE began around 2017, FYI. But here's the kicker, the one you keep ignoring, even when a translator is "translating" they are still leveraging a translation solution such as Trados, which handles machine translation for them through the use of a highly specific LLM trained either on data provided by the client or by the translator based on their previous translation work. And the sole reason for that, is that it's more time efficient for the translator.

Plus, when it comes to a large project, such as a significantly sized JRPG, there will be several translators working on the same material. Again, machine translation is leveraged with a glossary function to ensure that consistency of key words is maintained during the translation and subsequent editing processes.

I've been a professional translator working with various forms of machine translation (again, which have always been some form of LLM even before they were called that) since the early 2000s. I also post edit human translations, and can easily tell when someone has either left a machine translation unedited or not been capable of translating it themselves. Used to have one HUMAN translator who could never tell the difference between "sauce" and "source" when it came to translating user manuals.

[Ben Lubar]

Początkowo opublikowane przez Chika Ogiue:

Początkowo opublikowane przez Ben Lubar:

I've played many, many games that were translated into many, many languages, and they all used this technology called "a human" to translate them because it's cheaper than having to hire someone to go over all the wrong translations and fix them.

And that HUMAN was using a form of LLM because that's how we've translate these days. See, you've got it the wrong way round. It is cheaper to run machine translation -- which is essentially LLMs -- and then pay the PMTE rate which is LOWER than the translation rate. How low will depend on how the PMTE is being billed. If it's per source word (or character, depending on language) then that rate will be LOWER then the translators' translation rate. If it's paid by hour, it's significantly lower.

For example, a Japanese to English translator will typically charge between 6 to 12 JPY per source character to translate from Japanese to English. Their PMTE rate will typically be 4 to 10 JPY. Most agencies that hire for this have even lower rates for translators, as low as 1 JPY for translation and 0.80 JPY for PMTE. The shift from translation to PMTE began around 2017, FYI. But here's the kicker, the one you keep ignoring, even when a translator is "translating" they are still leveraging a translation solution such as Trados, which handles machine translation for them through the use of a highly specific LLM trained either on data provided by the client or by the translator based on their previous translation work. And the sole reason for that, is that it's more time efficient for the translator.

I've been a professional translator working with various forms of machine translation (again, which have always been some form of LLM even before they were called that) since the early 2000s. I also post edit human translations, and can easily tell when someone has either left a machine translation unedited or not been capable of translating it themselves. Used to have one HUMAN translator who could never tell the difference between "sauce" and "source" when it came to translating user manuals.

Trados is not an LLM. There are many applications of natural language processing that have nothing to do with LLMs.

https://www.youtube.com/shorts/KHEtJUlpqcg

An LLM is a very specific application of natural language processing and neural networks that encodes facts about words into a high dimensional space and then acts as a very sophisticated Markov chain, predicting what the next word in a document could be.

Even when ChatGPT translates text, it is not using the large language model to do so. It detects that someone is asking about a translation and then calls out to a different system to translate the text. Just like how the large language model doesn't generate images or search a database of websites. That information is provided to it as part of the document it is finding the next words for.

[Chika Ogiue]

Początkowo opublikowane przez Ben Lubar:

Trados is not an LLM.

It's software that leverages a form of LLM. What part of this is tripping you up? LLMs are not limited to only one function, nor do they exist in only one form.

Perhaps you'll listen to another solution spelling it out for non-professional translators:

DeepL’s proprietary neural networks and LLMs have been developed with AI pioneers and linguistics experts to grasp the nuance of sentence meaning.

FYI, DeepL is about as amazing at J <> E translation as Google Translate. That is to say, it's about a mid JLPT 2 level.

[Ben Lubar]

Początkowo opublikowane przez Chika Ogiue:

Początkowo opublikowane przez Ben Lubar:

Trados is not an LLM.

It's software that leverages a form of LLM. What part of this is tripping you up? LLMs are not limited to only one function, nor do they exist in only one form.

Did you not understand the video? What you are describing is not a large language model.

[crunchyfrog]

Początkowo opublikowane przez Ben Lubar:

Początkowo opublikowane przez Chika Ogiue:

And that HUMAN was using a form of LLM because that's how we've translate these days. See, you've got it the wrong way round. It is cheaper to run machine translation -- which is essentially LLMs -- and then pay the PMTE rate which is LOWER than the translation rate. How low will depend on how the PMTE is being billed. If it's per source word (or character, depending on language) then that rate will be LOWER then the translators' translation rate. If it's paid by hour, it's significantly lower.

For example, a Japanese to English translator will typically charge between 6 to 12 JPY per source character to translate from Japanese to English. Their PMTE rate will typically be 4 to 10 JPY. Most agencies that hire for this have even lower rates for translators, as low as 1 JPY for translation and 0.80 JPY for PMTE. The shift from translation to PMTE began around 2017, FYI. But here's the kicker, the one you keep ignoring, even when a translator is "translating" they are still leveraging a translation solution such as Trados, which handles machine translation for them through the use of a highly specific LLM trained either on data provided by the client or by the translator based on their previous translation work. And the sole reason for that, is that it's more time efficient for the translator.

I've been a professional translator working with various forms of machine translation (again, which have always been some form of LLM even before they were called that) since the early 2000s. I also post edit human translations, and can easily tell when someone has either left a machine translation unedited or not been capable of translating it themselves. Used to have one HUMAN translator who could never tell the difference between "sauce" and "source" when it came to translating user manuals.

Trados is not an LLM. There are many applications of natural language processing that have nothing to do with LLMs.

https://www.youtube.com/shorts/KHEtJUlpqcg

An LLM is a very specific application of natural language processing and neural networks that encodes facts about words into a high dimensional space and then acts as a very sophisticated Markov chain, predicting what the next word in a document could be.

Even when ChatGPT translates text, it is not using the large language model to do so. It detects that someone is asking about a translation and then calls out to a different system to translate the text. Just like how the large language model doesn't generate images or search a database of websites. That information is provided to it as part of the document it is finding the next words for.

It really doesn't matter whether it is or not.

These are dead ducks. You can't get over the simple inherent problems that such AI has and likely can never solve. There is absolutely no way for HUMANS to reliably tell the difference between satire and fact to 100% or even close. You have to have an EXTREME amount of knowledge, critical thinking, and so on you have to factor in.

And AI just isn't going to be able to do that.

AI's great for niche little narrow jobs, but inserting it into things like language will never likely be useful.

[Chika Ogiue]

Początkowo opublikowane przez Ben Lubar:

Did you not understand the video? What you are describing is not a large language model.

Yes, I get you don't understand the nuts and bolts of the translation industry, so you can stop trying to dev-splain it to a professional who has worked in it for decades.

I'm also sorry you don't understand that LLMs are not limited to one kind or one use, or that they have played a role in translation for far longer than the term LLM has existed in the public lexicon. But again, you're just going to have to deal with it.

[lx]

i played a mobile game once, the chat in there is using a translator and it makes it pretty entertaining. its good that it also have emoticons :D