Yee Cheng Chin

This one has been cooking silently in the background for a while now, but for #GDC2026 I'll be joining my industry peers Marcin Paczynski from GOG and Larry Kuperman from Nightdive Studios to combine our unique perspectives on how Game Preservation makes pretty good business sense! To quote the session description: > "Once seen mainly as a cultural responsibility, game preservation is now proving to be a smart business move. In this session, three experienced professionals working in remastering, platform strategy, and legacy content will each deliver focused micro-talks. Together, they'll explore how retro games can generate long-term revenue, extend the life of intellectual property, and reduce risk for both publishers and platforms."

72

5 Comments

As mentioned before, I have been engaged in some mad science experiments in the lab, mostly embedding models in Unreal Engine to drive completely unscripted behaviors and interactions by actors in a game. I moved from 7b (7 Billion) parameter models to 0.5b (about half a billion) with no loss in quality. Well, a researcher at Samsung has produced a paper and code for a 7 MILLION (yes, million, with an M) micro model that claims to beat massive foundation models like Gemini 2.5 pro on ARC-AGI test scores. Less is truly more if game developers want to embed these capabilities on consumer GPUs. Current models eat half of the GPU memory budget. Tiny models like this 7m use less space than a texture some artists will add to a character. There is even a direct correlation with size and speed, even with more compute dedicated to recursion and iteration with this approach. The possibilities are promising. Of course, from a technical perspective, this is exciting. As a matter of a successful business focused on crafting delight, I am much happier to sacrifice a texture or two for unbefore seen capabilities that can manifest on low-end consumer hardware for a larger market and dramatically advancing the state of the art to deliver fun. Imagine as a writer or designer focused on world-building, you are more of a "Dungeon Master" specifying motives and capabilities of characters and narrarive rather than hard-coding actions, dialog and consequences to bring your vision to players! No two play-throughs would be identical, but true to the vision you want to bring to life in your world! It is recent, so we will see if "too good to be true" applies, but it is worth keeping an eye on. Paper and code repo are here: https://lnkd.in/g4CZZFJa

5

1 Comment