Nerd Notes, February

Hey there Nerds...it's a long one this month. I had a whole thing that I wanted to get into, then DeepSeek R1 dropped and I started playing around with it and didn't want to wait a month to share something interesting that was happening.

So, this month's Nerd Notes broadly covers 2 topics:

• Pacifism in games that are based on aggression
• Why reasoning models still struggle with Pokemon TCG.

Let's get into it.

Why I retired from Risk

I retired from the board game risk in my mid-twenties after playing what I still consider to be the single greatest game ever played, aka the one where I peacefully conquered and held Europe for the entire game.

I promise this is relevant to the rest of the newsletter, so let me set the scene: I was playing the “mission” version of Risk with 4 other friends, the version where every player draws a mission card at the beginning of the game that gives them a mission to complete in order to win (as compared with the traditional world domination battle royale). I was playing as the red army, and the card on my mission was to eliminate the yellow army. When the initial board position was established, however, I recognized that my army was already heavily concentrated in continental Europe, so I decided to play a different mission. Over the course of the first round of the game, I made a series of trades: I offered up a territory outside of Europe in exchange for receiving a territory inside Europe. By the end of the first turn, I hadn’t rolled an attacking dice at all and had full control of Europe.

For every subsequent turn, I made 2 moves: I placed my 7 reinforcement armies somewhere along the borders (2 armies for holding 7 countries in Europe, 5 armies for holding the whole continent), and I negotiated a trade in order to gain a territory card. The trade always involved a country along my border where another player had only 1 army. I’d ask them to give it to me with the promise that I’d promptly give it back when their turn came.

From time to time, I’d announce some breakthrough cultural achievement that was happening in pacifist Europe because we weren’t expending our resources on defense. We were opening classical music institutes, new art galleries, particle accelerators, and so on.

At first, all of the other players were perplexed and not sure what to make of it. They were (understandably) concerned that I was playing some sort of long game that involved sustained deceit, and eventually I’d have amassed some huge, world conquering force.

But I never attacked.

And on my last turn of the game, whoever had been in control of the yellow army was in dire straits. There was actually a possibility that another player would wipe them out, and I’d have completed my mission without doing anything to complete my mission. Instead, the yellow army was backed into the Middle East as their only territory held and with a small contingent of like 6 or 7 armies. I was sitting in Southern Europe with over 15 armies and territory cards I could have traded in for more. If I had chosen to attack, the odds were overwhelmingly in my favor that I would have wiped them out.

This is the thing, though, even though I had a card that said that my mission was to wipe out the yellow army, I had decided my mission was to peacefully conquer and hold Europe…

So that’s what I did, and when someone else completed their “official” mission on the next couple turns, I had the immense satisfaction of turning over my official mission card and having everyone see that I could have done it if I had wanted to.

There’s no way to top that, so I retired. (I came out of retirement only once, when Sarah & I were staying at a hotel in Uganda while we were working in South Sudan. We had a few days off, we were holed up in the hotel and doing not much aside from eating, reading, walking, and swimming. But they had a copy of Risk, and she wanted to play, so I agreed. After about 30 minutes it was obvious she was going to win, so I conceded…which was apparently not a thing that one did in her family, so she was furious. That game did nothing to convince me to stay unretired.)

Sander watch: Birmingham edition

While I’m not coming out of the woodwork to play Risk anytime soon, my favorite Pokemon TCG player popped up in Birmingham with a new, never before seen deck, and played it all the way to the top 16. I am talking, of course, of Sander Wojcik. The greatest pleasure I felt from that tournament was riding the train home from the airport, pulling up Twitter to see that he had posted the deck, and then working out with Tommy how the deck actually worked.

It was a real thing of beauty, and also something totally ephemeral - it probably doesn’t work with the changes in the format starting with the next tournament. Sander’s style is control: his goal is to trap something and prevent the other player from attacking until they run out of cards to draw. He threw in 2 fun wrinkles with this deck that we hadn’t seen before:

• the Black Kyurem ex is a card I haven’t seen anyone play, much less in a top 16 deck. It is there entirely for the matchup against the current “Best Deck in Format” - Regidrago VStar. It paralyzes the Regidrago, preventing it from attacking or retreating back to the bench.
• But there is another way the Regidrago could escape: it could use a trainer card to scoop up the active and bring in something from the bench…and that’s where Sander’s 2nd fun wrinkle comes into play: he plays a Milotic (not the Milotic ex that we were playing) which has an ability that prevents the other player from scooping up their active Pokemon. This combo by itself is already evil, but he twists the knife a little bit more by also playing the Radiant Alakazam, which lets him move 20 damage from one of his opponent’s Pokemon to another. What he’s doing is using that to reduce the amount of damage on the Regidrago, because Black Kyurem’s attack does actually do damage, but by moving damage away he can keep the Regidrago trapped even longer. It’s hilarious.

(Actual deck list here if you really want to go down the rabbit hole)

Sneak preview: what I’m bringing to EUIC

And all of this left me inspired. Sadly, my Tsareena deck just doesn’t really work anymore. As long as Gardevoir remains a strong deck, Tsareena just can’t work…so I needed something new for EUIC, and I was playing around with a few different decks, but none of them were scratching that itch for me.

But now I’ve decided: For EUIC, I’m going to be a Pokemon Pacifist. I’m building a control deck. And while some control decks these days are set up to be able to attack if necessary, I’m going to track 3 stats for EUIC:

• Number of knockouts I take (ideally 0)
• Number of times I could have taken a knockout but chose not to
• Total amount of damage dealt (one of the main cards I’m going to play has to do a small amount of damage sometimes)

I haven’t totally settled on the full 60 cards I’ll be rolling with, but it’ll be some variation of Pidgeot Control, using Mawile & Wellspring Mask Ogerpon ex (Probably similar to what James Kowalski - aka Alloutblitzle - just took to 3rd place at the San Antonio Regional.) I’m guessing it’ll result in a lot of ties, and for a pacifist that’s OK.

The Fortnite Pacifist

And in that spirit, I’m going to reshare this video. It’s been in the newsletter before, but it just fits in so nicely with this edition on pacifism in Games.

A moment when you know you’re too deep down the rabbit hole

Sitting in a pizza place in a mall in Birmingham, I look out into the food court and see someone standing 100m away with his back to us and say to the kids “Oh look, there’s Tord Reklev.” And I was right.

Something completely different: Putting DeepSeek R1 through the Pokemon Deck Challenge

My current white whale is getting an LLM to build a competitive Pokemon deck. All of the non-reasoning models fail because they're trained on a dataset that is based on rotated cards, and I just can't get them to not recommend those cards. Also, their strategies are pretty crap.

I decided to put DeepSeek R1 through the test because it's a reasoning model, and if you run it locally as compared to using their app, it shows the reasoning process it's going through.

I'm dropping the more interesting part of the interaction below. This follows after I asked it to build me a competitive deck based off of Charizard ex and went through multiple rounds of trying to redirect it. The simplest thing to say is that it just fails spectacularly, but in a completely different way from something like GPT4, Llama 3, or Claude Opus. You can take a look at the comparative transcripts below, but it's very interesting to me to see it trying to reason through a question that actually just requires pre-existing knowledge. It really wants to figure out what the "ex" addition denotes, and it has some reasonable but ultimately wrong ideas.

In the images below, you can see:

• first, DeepSeek R1
• then Llama v3.3
• and finally Claude Opus, about which a funny note: I think I've seen people playing this deck online and been bewildered by it. I now wonder if they were taking suggestions from an LLM. It's not a good deck.
• GPT4o, not pictured was so long winded that I couldn't even excerpt it here. Also, it was abundantly wrong.