It’s nice to see a company try and maintain the human component in a time when that’s getting thrown out the window in favor of mass generated trash. Love your writing style also.
Jun 30, 2023·edited Jun 30, 2023Liked by Nathan Brown
"For those of who you are understandably struggling to parse that word salad, Sentis enables you to generate game assets with machine learning, rather than paying a human to do it. As part of this, a host of ML content providers were added to the Unity Asset Store."
Given the limited amount of information available on Sentis I may be mistaken, but I believe Sentis isn't specifically for "generat[ing] game assets with machine learning". I believe you're mistaking that for existing tools such as Atlas.
Sentis, as I understand it, streamlines embedding ML models in the Unity runtime, for execution on the player's device, well after the developer would have paid anyone to create the assets in question. Of course there's nothing to stop you also utilising it to generate assets if you want to write the code for it but AFAIK Sentis is a general purpose ML framework, not an asset creation pipeline.
Applications of ML in game runtimes are not entirely novel and you would utilise them for things like speech to text transcription, behaviour of AI agents, procedural animation generation, and other such fairly innocuous but interesting things.
Kotaku's recent article on the Unity announcement makes a similar (I believe) false assumption.
Quite alright. Given how little Unity have actually explained about it, its entirely possible I'm wrong and it's a tool to make it easier to embed ML tools into the Unity editor runtime, but the language makes me think it's a very well meaning piece of inference engine scaffolding, and it's unfortunately been sandwiched in between their AI assistant chatbot and a bunch of adverts for generative ML tools from external providers.
There's an interesting conversation to be had about _why_ everyone has both assumed the worst, and any mention of generative AI in a production pipeline is so fantastically toxic, but I'm professionally involved in something that does lean in that direction so I'm going to take the coward's exit and not have it.
It’s nice to see a company try and maintain the human component in a time when that’s getting thrown out the window in favor of mass generated trash. Love your writing style also.
Thanks Michael!
Callback to Edge issue 1? Deep cut.
Honestly amazed it’s taken this long. (However I believe it was issue 3.)
I stand corrected!
"For those of who you are understandably struggling to parse that word salad, Sentis enables you to generate game assets with machine learning, rather than paying a human to do it. As part of this, a host of ML content providers were added to the Unity Asset Store."
Given the limited amount of information available on Sentis I may be mistaken, but I believe Sentis isn't specifically for "generat[ing] game assets with machine learning". I believe you're mistaking that for existing tools such as Atlas.
Sentis, as I understand it, streamlines embedding ML models in the Unity runtime, for execution on the player's device, well after the developer would have paid anyone to create the assets in question. Of course there's nothing to stop you also utilising it to generate assets if you want to write the code for it but AFAIK Sentis is a general purpose ML framework, not an asset creation pipeline.
Applications of ML in game runtimes are not entirely novel and you would utilise them for things like speech to text transcription, behaviour of AI agents, procedural animation generation, and other such fairly innocuous but interesting things.
Kotaku's recent article on the Unity announcement makes a similar (I believe) false assumption.
This is really useful, thank you. And apols for the error. Will reference this in one of next week’s editions if you don’t mind.
Quite alright. Given how little Unity have actually explained about it, its entirely possible I'm wrong and it's a tool to make it easier to embed ML tools into the Unity editor runtime, but the language makes me think it's a very well meaning piece of inference engine scaffolding, and it's unfortunately been sandwiched in between their AI assistant chatbot and a bunch of adverts for generative ML tools from external providers.
There's an interesting conversation to be had about _why_ everyone has both assumed the worst, and any mention of generative AI in a production pipeline is so fantastically toxic, but I'm professionally involved in something that does lean in that direction so I'm going to take the coward's exit and not have it.
Good stuff :)
Thanks Tony! Humbled to know you are reading!
GenML's output feels like Schröedinger's content: it somehow belongs to both everybody and nobody, all at the same time.