Joined March 2019
4 Photos and videos
fun project from last night, eroding images
55
what's funny about this is that video game prices are mostly set by social norms
The average person believes prices are set by social norms even in one of the most competitive marketplaces a human being has access to
69
technical writing is hard :(
1
77
> does your healthcare budget still matter in a world where all diseases are cured? Yes, obviously? Just because a cure exists doesn't mean you can afford it. We have a cure for TB, but it still kills millions each year due to lack of accessibility.
>Everyone outside that group feels like they can work their well-paying (but <$500k) job for their whole life and never get there. Never get where, though? Many of the people in this industry believe takeoff is imminent or already underway. Do these numbers matter in the world they increasingly seem to believe is on the near horizon? Prominent people now describe scenarios before 2040 that would have been called extremely hyperbolic even a couple of years ago. Or to be blunt, completely insane. $20m net worth. $50m. $200m. $20b. Do these numbers matter at all in those futures? Is there any difference between them? Does your healthcare budget matter in a world where all diseases and illnesses have been cured? Does it matter how much land you have, or how big your house is, in a future where we are terraforming planets or building megastructures? Does it matter if you have enough money to escape your job and retire if no one works? If all human work is done by Minds, or embodied Minds? No. I know a lot of people think everything I'm describing above is hype to boost IPOs. Or an absurd fantasy. But when you hear some of these CEOs, or other prominent people in the labs describing this level of change being imminent, they believe it. When Dario Amodei talks about 'a country of geniuses in a datacenter' he's not trying to sell Claude, he is being sincere. I'm saying some of the people being envied don't think their money matters either - or won't, soon. Certainly not in the same way it does now. People in this community increasingly give these futures credence, but not their second-order effects. They imagine a future based on the science fiction they grew up with, where human civilization stays broadly structurally and culturally the same; people have the same hopes and fears, concerns and circumstances, only we also have spaceships, ringworlds, Dyson spheres and Matrioshka Minds. But those stories were intentionally written that way, with that false asymmetrical grounding, so that the audience could relate to the world the characters were in. Otherwise no one buys the book. TAI means Transformative AI. And transformative change means ๐˜ต๐˜ณ๐˜ข๐˜ฏ๐˜ด๐˜ง๐˜ฐ๐˜ณ๐˜ฎ๐˜ข๐˜ต๐˜ช๐˜ท๐˜ฆ ๐˜ค๐˜ฉ๐˜ข๐˜ฏ๐˜จ๐˜ฆ. It means all the old structures, the values, the rules, are washed away by the wave, the storm, the tempest already at the door. It means we live in a world we cannot even predict from where we are standing now, because what lies beyond the horizon is different in wildly unpredictable ways. Because it is completely out of distribution from what we have experienced, or even imagined. It means this: ๐˜๐˜ถ๐˜ญ๐˜ญ ๐˜ง๐˜ข๐˜ต๐˜ฉ๐˜ฐ๐˜ฎ ๐˜ง๐˜ช๐˜ท๐˜ฆ ๐˜ต๐˜ฉ๐˜บ ๐˜ง๐˜ข๐˜ต๐˜ฉ๐˜ฆ๐˜ณ ๐˜ญ๐˜ช๐˜ฆ๐˜ด; ๐˜–๐˜ง ๐˜ฉ๐˜ช๐˜ด ๐˜ฃ๐˜ฐ๐˜ฏ๐˜ฆ๐˜ด ๐˜ข๐˜ณ๐˜ฆ ๐˜ค๐˜ฐ๐˜ณ๐˜ข๐˜ญ ๐˜ฎ๐˜ข๐˜ฅ๐˜ฆ; ๐˜›๐˜ฉ๐˜ฐ๐˜ด๐˜ฆ ๐˜ข๐˜ณ๐˜ฆ ๐˜ฑ๐˜ฆ๐˜ข๐˜ณ๐˜ญ๐˜ด ๐˜ต๐˜ฉ๐˜ข๐˜ต ๐˜ธ๐˜ฆ๐˜ณ๐˜ฆ ๐˜ฉ๐˜ช๐˜ด ๐˜ฆ๐˜บ๐˜ฆ๐˜ด: ๐˜•๐˜ฐ๐˜ต๐˜ฉ๐˜ช๐˜ฏ๐˜จ ๐˜ฐ๐˜ง ๐˜ฉ๐˜ช๐˜ฎ ๐˜ต๐˜ฉ๐˜ข๐˜ต ๐˜ฅ๐˜ฐ๐˜ต๐˜ฉ ๐˜ง๐˜ข๐˜ฅ๐˜ฆ, ๐˜‰๐˜ถ๐˜ต ๐˜ฅ๐˜ฐ๐˜ต๐˜ฉ ๐˜ด๐˜ถ๐˜ง๐˜ง๐˜ฆ๐˜ณ ๐˜ข ๐˜ด๐˜ฆ๐˜ข-๐˜ค๐˜ฉ๐˜ข๐˜ฏ๐˜จ๐˜ฆ ๐˜๐˜ฏ๐˜ต๐˜ฐ ๐˜ด๐˜ฐ๐˜ฎ๐˜ฆ๐˜ต๐˜ฉ๐˜ช๐˜ฏ๐˜จ ๐˜ณ๐˜ช๐˜ค๐˜ฉ ๐˜ข๐˜ฏ๐˜ฅ ๐˜ด๐˜ต๐˜ณ๐˜ข๐˜ฏ๐˜จ๐˜ฆ.
1
152
Balatro TUI Balatro AI. Got it playing the first round flawlessly! (80% win in 1-shot) All it took was a 50M parameter neural net and 35 million analyzed positions.
1
1
91
More info for the curious: PPO nn has 3 parts: - trunk (resnet, 6 blocks, 1536 width, 2:1 ratio, ln, gelu) - linear layer to value head - linear layer to 512 width vec, added to embed vec for each move, passed through 1 resnet block, then gelu -> linear to get logit.
1
77
Also crazy that I can just leave my laptop running overnight and do more then 10^16 FLOPs
53
Technical writing makes you realize the huge gap the frontier of a field and the background most people have. I recently did a writeup on optimizing kernels for Perlin Noise, and I had to gloss over so much, while also assuming the reader had ~CS degree level knowledge (1/2)
May 11
no, goddamnit, feynman was just wrong, some ideas are genuinely irreducibly complicated. i know mathematics it would take literally years to explain to a layman, which is literally why graduate school exists as an institution, to do that high-level mathematics takes place in a very sophisticated language where every concept is defined in terms of other concepts which are defined in terms of other concepts etc. etc. etc. and fully unraveling all of these concepts back to what a layman would be familiar with takes potentially thousands of pages depending on the subject. and i'm not talking about stuff with no applications, this is the level of effort it would take to fully explain all the math that goes into the standard model of particle physics, eg other subjects are also like this! you might question whether some of these subjects are mostly fake, towers of theory disconnected from reality, but that's a different conversation entirely. the towers of theory actually do exist and require actual effort to climb
1
95
I could write a whole article on - What Perlin noise is for - How it works - How it's used - The math behind it - Hash function design - SIMD - Register pressure - floating-point bit hacking All of which are important! And each have their own knowledge dependencies!
61
My hope is that it is becoming a personalized feed, and it shows these high WL games because it's assuming a default user. So people interested in indie games will see indie games. Still, it's definitely a bit spooky to see this change.
So apparently Steam is testing a new Beta UI that changes how Popular Upcoming work. The current Popular Upcoming is sorted chronologically. However the new one seems to show games with massive wishlist which screws us small devs with limited budgets! Are we cooked guys?
63
this is awesome, but I always get a bit sad when I hear things like this, since I have severe ADHD and stimulants don't help me at all :/ why can't it be this easy
It is genuinely kind of insane how much stimulants fix not just my ADHD, but nearly every other corollary mental health issue I have. My anxiety isnโ€™t as present because it feels like I can act. My depression isnโ€™t as present because it feels like I can act
89
While I personally don't like meta progression, most people prefer it. Losing is demoralizing if no tangible progress is made. Games like Balatro and STS have navigated this by making meta progression that expands content, without making the game easier
Meta progression will always be a blight on video game design. I want to beat a game because I got better at it, not because the game got easier over time.
142
Deepseek v4 makes you really appreciate how difficult the 1M token context engineering problem is. They had to reinvent attention like 4 different times (MLA -> DSA -> CSA HCA) just to make inference affordable.
1
75
I just trained an AI to play the first round of Balatro, it turns out that the best strategy is usually to go for high straights (not flushing surprisingly) and this is the optimal breakdown of outcomes:
1
66
To clarify, this is on gold stake difficulty with decks that don't modify the first round (yellow, ghost, nebula, zodiac)
47
I think Factorio should honestly be on this list, given that it birthed what is now one of the most popular indie genres
For my money, if we're talking influence and impact on the industry, the design space and the culture, it's these four.
165
Most Perlin Noise implementations suck, I made one that doesn't: milesoetzel.substack.com/pubโ€ฆ

75
Miles retweeted
the secret to indie gamedev (no one will tell you): just make a fucking banger. everything else is survivor mumbo-jumbo
11
20
249
11,077
5 Nov 2019
Its crazy to think that every game in the "computers still lose to humans" category of this xkcd comic from 2013 has now seen computer beat the best players: Starcraft, Poker, Arimaa, and Go. imgs.xkcd.com/comics/game_aiโ€ฆ