LLM powered steam search engine
I think many of us scroll through steam to find that new game to play. I believe valve can use a llm a predictive search engine for better recommendations. Valve can come up with a new "description rubric" in order to properly match games. this can also improve developers knowing what direction to take their games
< >
Showing 1-5 of 5 comments
Ettanin 18 hours ago 
So you want Valve to rent ChatGPT or similar to allow users to dynamically find games based on a LLM prompt, burn money with virtually no hope of any return to provide this service free of charge to millions of users monthly out of Valve's own pockets and on top endanger the user's privacy to third parties?

Valve can't afford to build their own datacenters to have their own dedicated model for such a task, so renting is the only option. Valve is not the size of Google, Meta, Amazon, Microsoft or nVidia.

Do you really want Valve to sell our data to such services without our consent?

Because the big ones already burn billions upon billions of dollars for the insatiability of their LLMs and won't offer their services for free and Valve will also not foot the massive bill out of the goodness of their hearts.
Just.... ask a chatbot about videogame? Why do Valve need to own one for you to use it. Doesn't matter.
I'm not sure why you need a version of autocomplete that takes gigabytes of memory to run in order to search for games on a website that is already searchable on every search engine.
Originally posted by Ben Lubar:
I'm not sure why you need a version of autocomplete that takes gigabytes of memory to run in order to search for games on a website that is already searchable on every search engine.
I'm not sure why you need to post your obtuse non-answer just because you lack the OP's vision.

simple text search isn't the same as using LLM "comprehension" (such as it is) to get a little deeper understanding of the subject, or to draw correlations that are way too tedious using trad search (e.g. would take tons of searches instead of just getting a summarized answer from LLM.)

all that said, I agree with Lance that general purpose chatbots should be sufficient for this. i mean, they are often stupid af but anything on a smaller scale (i.e. valve built in-house) is only going to be stupider.

afa etta goes, dude is just a hater of ideas, thinking they know stuff, posting on threads that shouldn't concern them. like the straw man above. but i'll just throw out there, there is plenty of subsidized LLM going on in the world. (i.e. appearing "free" to the user). they just throttle you down when you're over the free budget. and depending on the service you use, privacy can be a feature. (e.g. MS Copilot actually licenses other models and offers them to MS customers with their own added privacy that goes far beyond what the native providers offer.)

and as for valve, who knows wtf they are doing? they haven't had a good idea in ever. steam machines lol OK. we can't really count on them for anything, since forever, and ppl should stop acting like we can. their last good idea (that they actually pulled off) was TF2. that's like 20 years ago.
Originally posted by Ettanin:
So you want Valve to rent ChatGPT or similar to allow users to dynamically find games based on a LLM prompt, burn money with virtually no hope of any return to provide this service free of charge to millions of users monthly out of Valve's own pockets and on top endanger the user's privacy to third parties?

Valve can't afford to build their own datacenters to have their own dedicated model for such a task, so renting is the only option. Valve is not the size of Google, Meta, Amazon, Microsoft or nVidia.

Do you really want Valve to sell our data to such services without our consent?

Because the big ones already burn billions upon billions of dollars for the insatiability of their LLMs and won't offer their services for free and Valve will also not foot the massive bill out of the goodness of their hearts.
As if they weren't already selling them...
< >
Showing 1-5 of 5 comments
Per page: 1530 50