Valve has launched "Steam Labs", a place where Valve will show off new experiments

So… what's there left for Valve to surprise us :P? (I like it so far)

*sigh* I got really excited for a moment about the Interactive Recommender, but there doesn't appear to be a manual override for the included games, which makes it useless for me. For example, two of my top five by playtime are clickers and I am *done* with that genre. What would be useful is if I could exclude any game I have hidden (or uninstalled, for free-to-plays).

Also doesn't look like it can be filtered by OS? I use Steamplay, but it would be nice to be able to look at the Native Linux stuff first.

I can't decide whether I think that recommender thing is creepy, but probably a good idea, or probably a good idea, but creepy.

Ah yes, Dragons, that classical game genre.

I quite like the 6-second trailers. Better than regular videos, that I never watch, and they could make for an interesting discovery queue or something (though I would like to have some text to help as well).

The machine learning part is... not ideal. I have the opinion that, if my personal data is going to be used to generate recommendations I should be the one controlling that. Which means it should run locally, with no one else having my data, and I should be able to change the parameters as I wish - granularly remove any data I don't want to be used, filter what games can be recommended (by price, OS, DRM and etc), and so on.

I like the idea of Steam Labs. My suggestion: broadcasting from Linux.

Cool.

But can I have a new Valve ARG?
Please.

QuoteThe first three experiments

So they CAN count to three...

Give me a call when they try the "use humans to curate the store so it's not brimming with garbage constantly" experiment, that's the one that I'm still really pining for.

Interesting stuff, I liked all of it!
The Quad video thing was my favorite.

A problem with the recommendations presented by the ML feature "Interactive Recommender" is that some of my best gaming experiences have been from short and focused interactive experiences which didn't take as much time to go through as some of my top games by hours played. Seems these games are not treated fairly by this algorithm. The recommendations makes sense in comparison to my most played games, but most hours played does not always mean most compelling game, and some of those most played games I'm actually done with and I'm burned out on those genres.

I think a partial solution could be to look at the average playtime for the entire user-base for a game, and then see how my time compares to that of the average user. If I spent more time than average, then I probably found it compelling. Some games are intentionally short, others can be huge time sinks. By looking at a ratio compared to an average instead of absolute playtime we can ensure the former category isn't treated as unfairly as it currently seems to be.

Edit: Turns out they are already doing something like this :)


Last edited by Brisse on 12 July 2019 at 10:32 am UTC

Quoting: BrisseA problem with the recommendations presented by the ML feature "Interactive Recommender" is that some of my best gaming experiences have been from short and focused interactive experiences which didn't take as much time to go through as some of my top games by hours played. Seems these games are not treated fairly by this algorithm. The recommendations makes sense in comparison to my most played games, but most hours played does not always mean most compelling game, and some of those most played games I'm actually done with and I'm burned out on those genres.

I think a partial solution could be to look at the average playtime for the entire user-base for a game, and then see how my time compares to that of the average user. If I spent more time than average, then I probably found it compelling. Some games are intentionally short, others can be huge time sinks. By looking at a ratio compared to an average instead of absolute playtime we can ensure the former category isn't treated as unfairly as it currently seems to be.

Hmmmm . . . but what if everyone, or even just most people, who ever bought a game thought it kind of sucked and didn't put much time into it? I see the problem you're trying to deal with but I'm not sure your solution does the trick either, because it would give a boost to low-quality games.
Not sure there's a simple rule that would handle the issue. But isn't it supposed to be a learny thing? So if so it shouldn't be working by a simple rule, it should be refining as it gets used.

Well, the recommender got me down to a T. Added a few to the wishlist.

Quoting: BrisseA problem with the recommendations presented by the ML feature "Interactive Recommender" is that some of my best gaming experiences have been from short and focused interactive experiences which didn't take as much time to go through as some of my top games by hours played. Seems these games are not treated fairly by this algorithm. The recommendations makes sense in comparison to my most played games, but most hours played does not always mean most compelling game, and some of those most played games I'm actually done with and I'm burned out on those genres.

I think a partial solution could be to look at the average playtime for the entire user-base for a game, and then see how my time compares to that of the average user. If I spent more time than average, then I probably found it compelling. Some games are intentionally short, others can be huge time sinks. By looking at a ratio compared to an average instead of absolute playtime we can ensure the former category isn't treated as unfairly as it currently seems to be.

Yeah, using time played as a metric for preference is deeply flawed. But I don't think looking at averages instead is a solution, for some of the same reasons: maybe you put more hours than average, but it doesn't mean the experience is better than that from another game where you put less hours than average.

Ultimately, the ideal metric is asking users "do you want more games like this or not".

Edit: it looks like they already normalize for time played like you suggested.


Last edited by eldaking on 11 July 2019 at 11:05 pm UTC

I like the idea of letting users get a look at Steam's "works-in-progress". Anything a company can do to give users insight and input into their services is a good thing. Although each of the three labs seems to be a rehashing of "Your Discovery Queue", I rather like the "Automatic Show Experiment" format. I can see myself going back to this every few weeks to get a quick perspective on what is out there in the Steam Universe. I cannot say I ever find myself wanting to review my "Your Discovery Queue".

Quoting: eldaking

Quoting: BrisseA problem with the recommendations presented by the ML feature "Interactive Recommender" is that some of my best gaming experiences have been from short and focused interactive experiences which didn't take as much time to go through as some of my top games by hours played. Seems these games are not treated fairly by this algorithm. The recommendations makes sense in comparison to my most played games, but most hours played does not always mean most compelling game, and some of those most played games I'm actually done with and I'm burned out on those genres.

I think a partial solution could be to look at the average playtime for the entire user-base for a game, and then see how my time compares to that of the average user. If I spent more time than average, then I probably found it compelling. Some games are intentionally short, others can be huge time sinks. By looking at a ratio compared to an average instead of absolute playtime we can ensure the former category isn't treated as unfairly as it currently seems to be.

Yeah, using time played as a metric for preference is deeply flawed. But I don't think looking at averages instead is a solution, for some of the same reasons: maybe you put more hours than average, but it doesn't mean the experience is better than that from another game where you put less hours than average.

Ultimately, the ideal metric is asking users "do you want more games like this or not".

Yes, letting the users interact with the inputs could help a lot. If there was a checkbox next to the top 50 list of games used as inputs then I would definitely uncheck some of my most played because I feel like they are no longer relevant to my current preferences. I've had my Steam account since august 2004, so it's almost 15 years old. Mostly played Counter Strike and HL back then. Recent years not so much. I guess some people change over time :)

the single 6sec trailer looks ok-ish (better then the thumbnail shit they presented on the store page during the summersale) - if they find a way to load the whole page faster on a slow connection
the video thingy is a direct: nope! never!
and they recommender thingy would be oki-sh if it did replace the current recommendation lists with a single page solution
Also where is my Linux/steamplay indicator on any of the pages?

Here is the big but: But why do I feel SteamDB becomming the only sane and fast option to navigate the shop?

Age of Wonders 3 is free to keep on Steam currently (11-15th of July) get it it a great game


Last edited by Schattenspiegel on 11 July 2019 at 11:32 pm UTC

Quoting: Luke_NukemWell, the recommender got me down to a T. Added a few to the wishlist.

Yep. I share the reservations others have expressed, yet it does seem to work surprisingly well. It isn't to know that I've already played GTAV and Just Cause 2 to death on the XBox 360, and 200 hours or so of The Witcher 3 from GoG. Mind you, it should know I'm running Linux and JC3 (its top pick by default) doesn't work under Proton. But it's certainly a game that'd be near the top of my own list if it worked. And if I fiddle with the age controls, restricting it to games released more recently, it comes up with Red Faction: Guerrila Remarstered, the original of which is definitely one of my top 5 360 games.

That's pretty impressive.


Last edited by Dunc on 12 July 2019 at 1:26 am UTC

I gonna say I'm fall in love with "Automatic Show", it's have very huge potential. The best alternatives to watch the games' gameplay/trailers and at the same time avoid/evading some bias youtubers/streamers/news/shill-curator bullshit-ness.

Quoting: Duncthe original of which is definitely one of my top 5 360 games.

Wow, that's a lot of games. :D