faithb74•24d agoProlific Poster

I finally listened to that guy at the gaming cafe who said local LLMs are overhyped for practical use

So back in February, I was at a small game dev meetup in Austin and this older developer kept telling me that running a local LLM like Llama 2 on my laptop would be more trouble than it's worth for generating NPC dialog. I thought he was just being a gatekeeper, you know? But after spending 3 weekends trying to get Ollama to work reliably with my Unity project (and crashing my system twice during a demo), I finally see he was right. The output was fine for single lines but totally inconsistent for longer conversations, and the setup time was brutal. Meanwhile I could have just used a simple cloud API for like $15 a month and gotten way better results. Has anyone else found that local models just aren't ready for real game dev workflows, or am I missing some secret trick?

3 comments

3 Comments

anthony88324d ago

Same thing happened to me with Stable Diffusion in Unity.

the_amy24d ago

anthony883 probably just had a bad setup, because AI tools work fine in Unity for most people.

skyler_craig21d ago

nah i gotta hard disagree here. i've tried multiple AI tools in Unity on completely fresh projects and they all had the same issues with lag and crashing. @anthony883's problem is actually super common from what I've seen on the forums. the "it works fine for me" argument always comes up but it ignores that the workflow is just not optimized for realtime stuff. most of these AI tools are built for offline rendering or generating assets outside of Unity, not for running inside the editor. people who say it works fine probably aren't doing anything complex with them.