No-One Knows How Anything Works Any More

It's Friday. Let's grab a bag.

Pre-Alphas Are Like Busses

You wait weeks for an invite to a pre-alpha then two turn up at once... 

I signed up for Star Reach's testing program when it was first announced. I wishlisted the game on Steam at the same time. A little later I also received a couple of "Invite Your Pals" emails from people already in the pre-alpha. Long story short, I ended up with three registrations for testing, only one of which was for the email account linked to the Steam account where I'd wishlisted the game.

Despite all of that, I didn't get an invite to an actual test even though, for quite a while, Playable Worlds were sending me emails every few days, mostly asking me to send invites to friends so they could join the waitlist too and asking me to wishlist the game, even though I already had. I presume there's no visibility for developers to cross-reference which emails belong to Steam accounts. I guess that would kick up some privacy issues especially in the EU.

I wasn't about to create more Steam accounts just to wishlist Stars Reach on all three emails. I figured if I got an invite on any of them, I'd I'd just install the client through the Steam account I already had and worry about it later. Or more likely never.

After a while the emails dried up. I hadn't heard anything from Playable Worlds for a couple of weeks until yesterday, when I got one telling me about a half-hour video from a conference in which Raph Koster, with the occasional help of a very uncomfortable guy from AWS who's clearly not used to speaking in public, explains "how Playable Worlds is leveraging Unity Engine on Amazon Web Services (AWS) to embrace simulation for creating online worlds." 

I have it running in the background as I write this and frankly I'm not getting much out of it but then I'm not really listening and it's not particularly my area of interest. I'd link or embed it but it seems to be private. I don't think there's an NDA on the testing itself but there may be on stuff like this so let's forget I mentioned it.

I was a lot more interested in the next email that arrived, anyway. That was the one I saw as soon as I logged in this morning. It told me I'd been invited to join the testing program. On two of the three accounts as it turned out, although I didn't spot that right away. 

As well as providing the necessary code for me to download the game, it asked me to join Discord, which would have been fine, only my Discord account is tied to a different email than my Steam account so I had to make a new one and now I have two. This is my last quarter of a century playing MMORPGs in a nutshell. You would not believe the number of emails and accounts I have accrued in twenty-five years, all tangled together like the leads at the back of my PC, which I also can't keep separate.

Now that's settled, I'm looking forward to getting a look at the game itself, although not as much as I was before I read all of Wilhelm's posts on it. His reports make it very clear this is real testing, not just marketing, as indeed do the emails from Playable Worlds. I can't say that encourages me. My inclination to do unpaid testing work of this sort isn't as strong as it would have been twenty years ago. When I sign up for tests these days, mostly I just want to get some blog posts out of the experience. 

That said, if and when I play, I will do my best to test what I'm asked to test and to give feedback as appropriate. I do have some sense of responsibility. The email mentioned an upcoming test but by the time I got to read it, it had already happened. Not that I would have been able to attend in any case. It started at two in the morning.

As soon as there's a test at a time I can make, I'll give it a go and no doubt write up what happened here. Stars Reach will get a post of its own then, I imagine, rather than having to share.

If You Hated The Movie...

This turned up in game coverage from the NME yesterday. I hadn't seen it anywhere else and I thought it was interesting so I bookmarked it to share and now here it is.

The gist, in case you didn't click, is that the disastrous Borderlands movie apparently led to a spike in sales for the Borderlands games. As the NME article makes clear, the film was a disaster on all fronts: critics hated it, fans hated it, audiences probably would have hated it but since hardly anyone went to see it there wasn't much of an audience to do anything.

The movie was a commercial and aesthetic disaster and yet in one of those "financial meetings" that get reported sometimes, where the suits tell the investors how something that looks terrible is somehow good for business, the CEO of Take-Two, Borderlands' publisher, said that "Even though the film was disappointing, it actually benefited our catalogue sales."

I don't exactly know what "catalogue sales" are but I imagine he means all the buzz around the film, deeply negative though it was, made a bunch of new people curious enough to go buy the games. It'll be that old saw about "No such thing as bad publicity" I guess.

As it happens, I already own some Borderland game or other thanks to Amazon Prime. I haven't played it, of course. Neither have I watched the movie, although I did think the trailer looked pretty good and said so when it came out. I still think it'll probably be better than its reputation suggests and I would quite like to see it to find out if I'm right but I have several thousand movies on my watchlist I'd rather see first so it could be a while before I put that theory to the test. 

 

What Are They Doing In There?

As I was leaving a comment on Tobold's blog this morning, to suggest he probably knew AI didn't work quite the way he was saying it did, it occured to me that maybe I didn't know how it worked either. I mean, I thought I knew but I wasn't sure. 

Since it's always embarrassing to be caught out telling someone they're wrong while being just as wrong yourself, I did a little research. I didn't ask an AI this time, not so much because it would have felt weird, given the topic, more because I just seem to have fallen out of the habit. 

I googled around, looked at a few articles, then came across this piece, appropriately enough at a Substack site called Understanding AI.org ,and it made some things I hadn't been sure about feel a lot clearer than they ever had been before. So I thought I'd share. 

The article is called "Large language models, explained with a minimum of math and jargon" and it does exactly what it promises, to the extent that when it says "a minimum" it means there is some math, not that there's none. And some jargon, too, but I like jargon.

I was able to follow it, even with the minimal math, and I learned a lot, the main thing being that no-one currently understands how LLMs work, not even the people who develop them and work with them. There are whole research projects dedicated to trying to figure out how the LLMs do what they do, a process that the authors of the article estimate will take "years—perhaps decades" to come to any kind of conclusion.

I am not going to attempt to paraphrase or precis the article. I could get an AI to do it. That would be amusing. But I won't because, as one of the two authors says in the introduction, it's "the result of two months of in-depth research" and it deserves to be read in full. It's very much worth the time. 

After reading it, I feel I now have a much clearer sense of what's happening when I ask an AI to do something for me. Well, an LLM, I should say. The terms are not synonymous.

I'm also going to be less fractious about using the term "AI" in reference to LLMs in future. I've very much been in the habit of making it quite clear I feel the term is a misnomer and that using it is misleading. After reading this piece I'm less sure that's as much of a linguistic hill worth dying on as I thought it was. 

They may not be conscious or self-aware but LLMs are capable of constructing a picture of the world in ways we can neither emulate nor understand. There's long been a widely recognized issue in scientific circles over rating or even acknowledging animal intelligence, an argument that of late has shifted to include plants and fungi. Machine intelligence is sure to join the list, if it hasn't already. 

It boils down to a problem of definition. Humans decide what constitutes intelligence and frame it in terms we understand but for the most part we either can't or won't look outside our own operating parameters. The chances of us recognizing different intelligences when they present, if they don't present as human, seems slim.

Agree or disagree, we're stuck with what we're calling AI now so we may as well try to learn how it works. It's interesting to think about, at least. I recommend the article to your attention.

And Now, A Song

And now, as is the tradition here, some music. I've spent much of the last month or so listening to Christmas songs in preparation for this years Advent Calendar so apart from the 100+ songs about snow and Santa I've bookmarked so far, pickings are pretty slim. Let's see... ah yes, this'll do...

Phone Booth - Telescreens

I missed The Strokes first time around. Nice to get a second chance. Surprised they were able to find a phone booth for the video. Maybe they built their own.

Takes a while to start. Stick with it.

I'm out.

*** Some notes on AI used in this post ***

Just that picture of a man in a white coat, produced by Flux Schnell via Nightcafe from a prompt originally reading "A very secretive computer doing something secret in a very secretive way. Line drawing. Color. 1950s.". I then evolved the result once, adding the phrase "Make the man look more puzzled.". I don't think he does.