Daniel Meirom (@danielfilms) profile picture

Film Maker and Visual Artist. Producer of https://xitter.gq/t.co/BQlYGxW6ha

Philadelphia, PA
Joined June 2009
Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)Photo by Daniel Meirom (@danielfilms)
Want to play poker but don 't have the time? You need to listen to @JenShahade ⬇️ Play for just $5 on PokerStars exclusively on @FanDuel. Hit the top multiplier for the $1m jackpot See psta.rs/SpinFieldGoalUS
1
6
10
840
Daniel Meirom retweeted
Hi
12
8
211
2,751
Daniel Meirom retweeted
A blue wave is coming! Happy Halloween 🌊🎃🗳️
11
7
1
221
5,566
Daniel Meirom retweeted
Don’t eat the Poisoned Pawn tonight! 👻♟️
5
17
1,154
Daniel Meirom retweeted
The Knight and the Seven of Hearts: Last week I saw an interesting tweet by writer Adam Zachary Wasserman, showing a chart of LLM and human choices when asked to “pick a number from one to ten", showing that LLMs pick seven every time. The caption read “This is the least surprising graph in the world if you understand how gradient descent works.” LLMs aren’t trying to give you a random result, but a likely result, and their training data contains lots of sevens. So that’s what you’ll get. Again and again. Unless you prompt it not to. Seven, as the chart shows, is also the heavy favorite “random choice” by humans who are picking a number one to ten. In chess master and film-maker Wesley Wang’s acclaimed and wildly popular short film nothing, but everything. (2023) he riffs on why. LLMs have picked up on this pattern so emphatically that they just pick seven every time when you ask something like “Pick a number one to ten” or “pick a random number one to ten.” It’s possible to break the pattern by asking over and over in the same chat, or with more specific prompts, like “Pick a number one to ten via randomness”, or “Generate a number 1 to 10 using randomness.” This varies the answers, often triggering an outside tool. But it does make you wonder. Where else does this effect pop up when you’re chatting or ideating with these LLMs? The answer is: everywhere. If I ask Claude or ChatGPT about their favorites in chess and poker, they also converge. “Pick a chess piece.” The Knight! “Pick a random chess piece.” The Knight! Ten times in a row each, Claude and ChatGPT selected the knight. Claude gave a little context into why it loves the horsey so much. “It’s the one piece that doesn’t move in straight lines, and the only one that can get somewhere nothing else can reach.” “Pick a card” was the seven of hearts or the seven of clubs, with the ten of diamonds being the only other contender in dozens of iterations. Claude again explained their choice, “The one that’s no one’s favorite, which is exactly why nobody puts you on it.” Responses like this show the danger of using LLMs “just for brainstorming.” What feels like a random, organic process may be leading us down a narrow path of cool sounding numbers and ideas. And by the same token, when using LLMs to write anything personal, from essays to cold pitches to thank you or recommendation letters, it may sound to a savvy recipient like the finger is jammed on the seven key. Cool number, but is it that good? Interlude: Solve a White to Move and Mate in Two with all knights I pointed out in Time Magazine that LLMs converge to popular chess openings like the Sicilian Defense, as well as aesthetic themes like the Smothered Mate. It’s become cliché to assert they favor certain words and punctuation marks. Lately, I have seen more subtle incarnations: influencers reciting popular themes and “fun facts”, with verbal tics and topics making it clear they’re reading off an LLM script. The strategy often works: many of those reels get thousands of reactions. There are no sources. Part of the point of short form media is to avoid all hint of a source. When I copy/paste my entire substack essays into Facebook or X posts, I can’t add the dozens of links I do here. So I point people to my stack for the sourcing, but it’s obvious: plenty won’t bother. For reels and social media, this is a feature, not a bug. There’s a connection between convergence and destroying IP. If we don’t have our own ideas, why would anyone own them? Luckily, you can train your divergence muscles. A simple game like Rock Paper Scissors uses those skills, as does the far more complex game of poker. But you can also start with extremely direct games, where the whole conceit is to predict others’ words or ideas, and then diverge. In one of my childhood favorites Scattergories, you pick a fruit, or a president, or a piece of furniture that the other players would not pick. For example, a fruit that starts with P. Plum, peach, pineapple: too obvious. What about pomegranate or plantain? Those could really win. But what if your opponent is thinking the same thing and everyone thinks “peach” is too obvious? Peach it is! Now you’re also playing P, for poker. I just learned about a cool game, Krillion, a form of Scattergories in which you win points the rarer your guess is, as you scuba dive deeper into the ocean. I like the mechanic. Sinking is usually bad, but here it’s good, an example of divergent thinking within the game. I played a few rounds, botching some answers—who knew Pinot Noir (name a type of wine) was THAT obvious, while crushing with Cortado (name a hot drink) and Sierra Leone (name a country with a coastline). As we are surrounded more and more with LLM generated text, images and ideas, the value of divergence will go up and up. Because as pretty as the chess knight and the seven of hearts are, picking what everyone else is picking is a losing strategy. Knight to f7 mate.
4
2
15
1,726
Bring your best game on Sunday. Our Main events are back again with a combined $80,000 in guarantees: 🔥$60,000 Sunday Dynasty, $100 buy-in, 6pm start 🔥$20,000 Sunday Shield. $30 buy-in, 6.30pm start @jenshahade
5
7
642
Daniel Meirom retweeted
jennifershahade.com/stack/wh… Why I’m supporting @CISalinasss’s @TheInnCenter poker tourney on Wrongful Convinction Day ⚖️
5
12
1,011
Daniel Meirom retweeted
Starting soon: Day two of @PokerStarsUSA September Dynasty Main. 95 players remain out of 860. GL to me and Go Birds!! #PokerStarsAmbassador
6
1
19
1,225
Daniel Meirom retweeted
Poker Ladies Weekend 2026: I’ll be doing an online masterclass for the London event and at the Philly event. 👋
There are poker tournaments taking place across 14 different countries, Oct 16-18, for the @wpf_Poker Ladies Weekend ♀️ And if you're in London you'll get to catch up with @MarleSpragg and @JenShahade. Details below, including your nearest event 👇 poker.org/latest-news/wpf-la…
1
14
1,201
Daniel Meirom retweeted
GL to everyone in tonight's @PokerStarsUSA events, which also includes a 1K with a 100K guarantee and a $50 250K guarantee.
Your Dynasty starts here! The $3.5m Guaranteed prize pool September Dynasty Series starts today with the first six events of the festival starting at 7.05pm Take a look at Event #3 at 8.05pm the $100 Kick-Off, $40,000 Guaranteed The festival includes a $15,000 series Leaderboard. Every cash from every event counts! See the full schedule at psta.rs/USDynastySeriesSched
1
8
1,305
Daniel Meirom retweeted
25 years ago I was in Brooklyn doing some homework outside my tiny Williamsburg apartment, because it was such a beautiful morning. When the planes hit I watched in horror from across the river. The subsequent hours and days are now a blur. But the feelings of grief and solidarity were so vivid, lasting for months. In New York, everyone had so much compassion for each other. But there were also tremors if a pin dropped. We were all so on edge. Like so many, I was never the same. This 9/11 hits differently for me. It is the first year, 4th grade, that my son is learning about September 11th in school. So I told him about my morning and the bravery of the first responders and Flight 93. Every 9/11, I think about the everyday heroes we lost far too soon. Grateful to be here to pass on those messages of hope and bravery to the next generation. Let’s roll. #neverforget 🇺🇸
2
5
64
2,382
Daniel Meirom retweeted
Won with 95o in the Maryland State Champs six max. @PokerStarsUSA Because of course! RIP Dolly Parton 💝
2
3
48
1,983
Daniel Meirom retweeted
We saw it in chess, are words next? #ai #chess
5
6
44
1,672
Daniel Meirom retweeted
Chess: the art of the possible!? Fabian and I discuss why stronger players look more at “empty squares” 💡
2
5
1
28
2,097
Daniel Meirom retweeted
Jennifer Shahade’s mom gave her the most dangerous poker pitch imaginable: “It’s free money for smart people.” Which is already a hilarious thing for your mom to say. Even better? Her mom was a chemistry professor…and a bookie. Thanks for stopping by to give us a glimpse into your life @JenShahade! Watch the full episode here: youtu.be/5iaqHmXs-2w
3
6
1
17
3,932
Daniel Meirom retweeted
Can you solve this stunning mate in two, while making baskets? 🏀♟️
1
1
27
1,753
Daniel Meirom retweeted
Via @ACMChess official Facebook page: “I retract the article I wrote about WGM Jennifer Shahade from American Chess Magazine issue #37, and apologize for and regret any harm it may have caused her. I continue to support the efforts of Jennifer Shahade and the players who came forward in their efforts to make chess safer for all." - Pete Tamburro
2
5
70
6,383
Daniel Meirom retweeted
I took the day off work and went to the park to listen to other authors speak about writing. One of them happened to be one of my favourite ones: @JenShahade 😍
The @bryantparknyc Reading Room Event was so magical. Books, salon atmosphere, perfect weather. Reminds me of why I loved living in NYC 💕
3
4
8
753
Daniel Meirom retweeted
A best-selling crime author, NYT obituary writer and chess champion walk into the park....and oh wait, this sounds like it may end badly. Come watch! @bryantparknyc Reading Room authors panel on Aug 12th (Wed) 12:30 PM with Casey Sherman and Sam Roberts Free
4
10
1,210