Thanks for playing! Yeah the relationships can be fun. For me today was pretty straightforward: I guessed “cow” (a word I often guess) and nearby words were “equines” and “clops” which obviously made me try “horse”. Getting from there to the target required a smart hint though since I hadn’t explored the map enough. The smart hint was “bun” and nearby was “braids” so that gave me the goal, “pony”.

I’m still trying to come up with rules for what makes a good word.

• It shouldn’t be too specific. I tried “hydra” because it was the wordle day recently, and it’s terrible. All the nearby words are from greek mythology and don’t really help make progress, you can only keep on guessing other random greek mythology terms.
• It shouldn’t be too general. E.g. for “old” you just get a bunch of words that have to do with old things. And then when you try “old” you feel stupid, not smart. Somehow the action “average of words” feels less satisfying than “sum of words”. So e.g. today “horse” would have been worse than “pony” since horse is more general and pony is more specific.
• It has to be a word that everyone knows.
• It should have multiple prominent meanings. E.g. “pony” refers to both an animal and also to hair. (obviously every word has multiple meanings, but e.g. for horse its main meaning dominates)
• Ideally there shouldn’t be too much of a pattern to the words. It’s fun to do “the” one day and “pony” the next because they have completely different paths leading to them. (I don’t think I actually have “the”, maybe it should be a future word…)

Currently I’m manually going through lists of the most common words and filtering aggressively, but maybe I can get an LLM to do this. Unfortunately these rules can only be discovered by trying more words and playing them.

I currently have words generated ahead of time until June or July. So there’s some time to figure out the algorithm.

Today, dog + tail = horse — this game is teaching me ways to think about semantic relationships.

This sounds awfully confident. Why would the government have to get rid of all of 50% of the workforce? If you produce more, why does that support fewer people? Why would monopolization and automation eliminate 50% of jobs even without AI? This seems like it would have happened in some country of the world already if true.

If AI (plus robots?) does replace 50% of workforce, avoiding social collapse would require the government to either get rid of all of them, turn them into slaves, or feed them for free and limit children (i.e. get rid of them, but slowly, filtering out the smart ones).

The interesting thing is that even if AI would not exist, monopolization (one operating system, one banking system, etc.) and automation (machinery farming, producing) could over time effectively eliminate close to similar amount of jobs, retaining for the working class just the essential ones: maintenance, healthcare and education.

The important part is how this transition happens, what happens with the eventually obsolete part of society. AI helps make this transition even more sudden, and as such, potentially more bloody.

Thanks for the comment, even if it probably disagrees with my view. I will say that many of these improvements help the models run more efficiently, using less power. See e.g. Karpathy posting on the same day as this blog post here:

https://x.com/karpathy/status/2017703360393318587

What used to take 7 days now takes three hours, using much less power in total. So while the concerns about power usage are valid, especially since people keep on scaling up, there’s also work to make these more efficient. Also see the point about slower scaling above. You can see this in how the prices for inference have gone down over time. That’s less power usage and cheaper hardware. The reason that data centers keep on scaling up is that there’s more demand.

I also wouldn’t focus so much on the problems. It’s clearly imperfect technology, but it can do really impressive things. Look at what it can do and what it’s useful for, not at what it can’t do. Can’t use a hammer to clean windows either, but that’s no problem.

I share the concerns about privacy violations (e.g. the only social medias I’m really active on is blogs and Youtube) and see many problems with technology, both AI and past technology, but I am still fundamentally very optimistic about the future. I’m hoping for a future like Iain Banks’ Culture, and I don’t think people working towards something like that are evil.

Twenty or so years ago unaccountable tech companies stole our privacy. Today every purchase, every online interaction, every physical movement and more are tracked. The public should have been appalled by an invasion of privacy like this. Coming within living memory of the Gestapo, and so recently after the Stasi in East Germany, we should have risen up and gotten rid of, permanently, these companies and the people in them who made the decisions.

Well now we have a similar choice. Do we want every piece of human ingenuity and creativity stolen by huge tech companies just so it can be regurgitated to us as slop? This doesn’t have to happen. Anyone working on this or supporting it is evil, a dark agent against humanity itself.

It doesn’t even work. It’s all based on something that doesn’t know how many ‘r’s are in ‘strawberry’, or how many sisters a girl in a family of one boy and three girls has. Marginal, incremental improvements are made with exponentially higher levels of computation and power usage. The gains the world is finally making with renewable energy are being wasted on this, and much much more. The inevitable result is plain to see.

This doesn’t have to happen, and it shouldn’t. The only reason it’s happening at all is because corporate executives want it as an excuse to cut jobs and wages.

However, the thing that always made Avalanche great was that culture outside of the office. Blowing off steam doing NYC things at NYC places always made the misery of the work more tolerable. Most of my fond memories of Avalanche during Jc4 are of that time spent with others outside of the office – even at work events like Coney Island or whatever bar the christmas parties occurred at. These connections are what I believe made the DLC work feel like a major shift(at least the ones I was a part of). And to me was a strong signal that the studio could have grown into something great with Contraband.

Except, as you alluded to many times in the main post, the NYC<–>Stockholm relationship continued to never work – I felt it as soon as I joined that project. I could ramble on about my perspective of Contraband, but that would become a book. In short, it was a game that should have been cancelled years ago. Some of the development practices that you have mentioned did improve quite a bit in central tech – but less so on the projects(not just Contraband). Many will probably disagree with me, but I thought the engine and tech suite as a whole has been going in the right direction for quite a while, and its the game teams that are incapable of making games, in any engine.

Anyway,
Post-covid Avalanche was never the same. The culture was dead. Moving to the office in k-town became a metaphor for ‘the scrappy studio that could’ turning into a small cog in a wannabe-corporate clock. Work events felt like work and celebrations felt forced. The only time it felt a shred like old-Avalanche was after we were shut down, like the ghost of JC manifested one last time.