Microsoft's AI will be inside Minecraft, and other Xbox, PC games | Hacker News Reader

Microsoft's AI will be inside Minecraft, and other Xbox, PC games

windowscentral.com

20 pointsblibble24 comments

> Players will be able to use natural language to ask questions like "How do I craft a sword?" and the Copilot will search your chests and inventories for the necessary materials, or guide you to them if you don't have them. It will also explain how to craft the item, and so on, eliminating the need to alt tab and read a website for Minecraft guides

"Inside" as in "LLM search engine for game tips"

We're over a year into the AI hype, and this is the practical application? A game-adjacent search engine? The next AI Winter is going to give us all frostbite.

I don't think you need to worry about this being "the" application, it's being shoved down our throats everywhere.

We are the application.

It takes a long time to educate the general population and I think we're JUST getting to the point where people are distinguishing between AI and LLMs. Thankfully this process is sped up by the current crop of executives shoving it into every nook and cranny available.

I'm not into Minecraft, but would absolutely love having something like this in Factorio, or better yet in Blender.

While I doubt you can cover all eventualities additions like this would indicate to me that there's room to improve in how you present information or teach the player within the game itself. I find it hard to believe an AI response will provide the best experience, and the wikis presence indicates most 'figure it out yourself' seems to have gone by the wayside for many players already so why not cut to the chase and have a straightforward crafting menu (and maybe a 'mystery mode' that leaves you to figure it out if you want to).

On a larger scale, it also seems to demonstrate the different motivations in playing, or the destination versus the journey. For me part of the reason I'd play Factorio is engaging with problem solving assuming the relevant breadcrumbs are there to follow or puzzle pieces to assemble. Following distinct instructions to goals is a different experience that I'd look to other games to provide.

I don't think this is the big deal use case for LLMs as much as Microsoft adding their shiny new thing into all the products they have, as in tradition.

This use case is definitely a solution in search of a problem though. Their stated example is something that can be done quickly and locally - for over a decade, mods could list down crafting/processing recipes, item uses and sometimes even highlight inventories with needed items. Everything else about Minecraft's undiscoverability could be solved by writing a comprehensive in-game guide and redoing the achievement system - something sensible instead of calling an LLM.

> Microsoft adding their shiny new thing into all the products they have, as in tradition.

The simplest explanation is usually the correct one.

That brings up another point: This particular LLM system will require custom integration by the game developer, development that could go to all those features you mentioned. That sounds like a lot of work for a dubious payoff for either the player or developer.

Were knowledge databases or Lisp based NLPs ever useful or as used as much as LLMs? I don’t really understand the AI Winter analogy but perhaps I don’t understand the history that well.

AI Winter is an extreme example of the hype cycle, and refers more to investor expectations than technical capability. In the historical case a lot of investment went into AI (Summer), and the disappointing performance caused a deep pullback in investment dollars for decades (Winter).

Another way to frame my question above: Is this feature a good demonstration of AI's value (present or future) in the eyes of Microsoft's investors?

MS originally included Solitaire in MS 3 to improve peoples computer skills with drag and drop.

https://en.wikipedia.org/wiki/Microsoft_Solitaire#History

I agree with you; Solitaire is superb software! My humbug is how phoned-in this use-case feels, especially given the hype behind AI over the last year. There is a lack of imagination for what LLMs bring to the table.

I suspect the reason is AI isn’t trustable for important things (yet?) so they can’t do the big cool things without getting sued out of existence when the AI destroys lives because of bad medical or investment or safety advice.

Yeah that's what I was kinda getting at, "lets put this somewhere people are going to interact with it and learn but the stakes are very low"

Why? To waste resources and electricity? This is Blockchain mania all over again.

Optimizing our workload the impress the stock purchasing bots that like seeing AI in press releases.

Maybe as a prototype? It's pretty low risk, worst case scenario they screw up some things on Minecraft and it's stupid.(Assuming it does not go off the reservation ala Sydney)

I could imagine using it as a source of training/experience on how to practically create a help bot without having to worry about anything that actually matters.

Since it is a video game they can also probably put in a lot of analytics on whether they are useful (did you follow the directions to make the diamond sword)

My thoughts process is that this is all less than 2 years since chatgpt hit, and while a bunch of people can see the edges of practical use cases, people are still trying to figure out how to productionize things.

Long term I integration with office/complex programs would be amazing, imagine being able to ask in natural language to fix the formatting in a particular way without having to spend 15 minutes looking it up. (Thinking back to when I originally did my resume formatting)

CIOs don't have strategy. They just notice what other companies are doing and scramble not to be left behind of new tech, whether that makes sense or not.

Opt-in. Please let it be opt-in

at most it will be opt out.

A nice idea to make it easier to navigate your way through the gane, but by basically baking in user made strategy guides they can make even more unexplained game mechanics as well as further encourage players into cookie cutter play methods.

The question of model collapse is interesting.

They suggest that AI will eat the game faq publishers, and it will because it will consume all their traffic, but there’s no end of Reddit users posting this data for free. I don’t see this leading to model collapse, but I do see this moving content behind subscription walls.

Seems like a value opportunity for Reddit and Microsoft, and a death knell to independent publishers.

People need to start make posts using ai. Maybe create browser extension that will autogenerate reply.

"How do I build a sorting system for unstackable items?"

Reply on news.ycombinator.com